CMMI® or Agile: Why Not Embrace Both! Hillel Glazer, Entinex, Inc. Jeff Dalton, Broadsword Solutions Corporation David Anderson, David J. Anderson & Associates, Inc. Mike Konrad, SEI Sandy Shrum, SEI
November 2008 TECHNICAL NOTE CMU/SEI-2008-TN-003 Software Engineering Process Management Unlimited distribution subject to the copyright.
This report was prepared for the SEI Administrative Agent ESC/XPK 5 Eglin Street Hanscom AFB, MA 01731-2100 The ideas and findings in this report should not be construed as an official DoD position. It is published in the interest of scientific and technical information exchange. This work is sponsored by the U.S. Department of Defense. The Software Engineering Institute is a federally funded research and development center sponsored by the U.S. Department of Defense. Copyright 2008 Carnegie Mellon University. NO WARRANTY THIS CARNEGIE MELLON UNIVERSITY AND SOFTWARE ENGINEERING INSTITUTE MATERIAL IS FURNISHED ON AN "AS-IS" BASIS. CARNEGIE MELLON UNIVERSITY MAKES NO WARRANTIES OF ANY KIND, EITHER EXPRESSED OR IMPLIED, AS TO ANY MATTER INCLUDING, BUT NOT LIMITED TO, WARRANTY OF FITNESS FOR PURPOSE OR MERCHANTABILITY, EXCLUSIVITY, OR RESULTS OBTAINED FROM USE OF THE MATERIAL. CARNEGIE MELLON UNIVERSITY DOES NOT MAKE ANY WARRANTY OF ANY KIND WITH RESPECT TO FREEDOM FROM PATENT, TRADEMARK, OR COPYRIGHT INFRINGEMENT. Use of any trademarks in this report is not intended in any way to infringe on the rights of the trademark holder. Internal use. Permission to reproduce this document and to prepare derivative works from this document for inter- nal use is granted, provided the copyright and "No Warranty" statements are included with all reproductions and derivative works. External use. Requests for permission to reproduce this document or prepare derivative works of this document for external and commercial use should be directed to firstname.lastname@example.org. This work was created in the performance of Federal Government Contract Number FA8721-05-C-0003 with Carnegie Mellon University for the operation of the Software Engineering Institute, a federally funded research and development center. The Government of the United States has a royalty-free government-purpose license to use, duplicate, or disclose the work, in whole or in part and in any manner, and to have or permit others to do so, for government purposes pursuant to the copyright license under the clause at 252.227-7013. For information about purchasing paper copies of SEI reports, please visit the publications section of our website (http://www.sei.cmu.edu/publications/pubweb.html).
Acknowledgments The authors thank the following individuals for their significant contributions to this report: Mike Phillips, Sabine Canditt, Noopur Davis, Bill Peterson, Mary Van Tyne, and Barbara White. We greatly appreciate the insights and efforts of all of these individuals. In addition, the authors would like to recognize those who attended Consultant’s Camp 2007. Ta- ble 1 in Section 10 of this report builds on work done in a session at that event.
SOFTWARE ENGINEERING INSTITUTE | iii
Abstract Agile development methods and CMMI (Capability Maturity Model® Integration) best practices are often perceived to be at odds with each other. This report clarifies why the discord need not exist and proposes that CMMI and Agile champions work toward deriving benefit from using both and exploit synergies that have the potential to dramatically improve business performance.
iv | CMU/SEI-2008-TN-003
1 Problem Definition Agile development methods and CMMI best practices are often perceived to be at odds with each other. If these perceptions or their causes are not resolved, we are likely to see more confusion and conflict as the adoption of each approach increases. In addition, each approach includes prin- ciples of good software development often overlooked but needed by the other approach thus be- ing knowledgeable in both is important to project success. In the long term, this discordant situa- tion is not healthy for the software engineering profession. Why the discord between Agile and CMMI camps? The purpose of this report is to clarify why the discord need not exist and to ask for your help in making the software development communi- ty aware that, when properly used together, CMMI and Agile can dramatically improve perfor- mance. We believe there are two primary reasons for the discord between the Agile and CMMI camps: 1. Early adopters of both CMMI and Agile methods represent extreme examples of their soft- ware development paradigms. Early CMMI adopters were developers of large-scale, risk- averse, mission-critical systems, often with high levels of management oversight and hierar- chical governance; whereas the early adopters of Agile methods generally focused on small- er, single-team development projects with volatile requirements in a software-only environ- ment. These two extremes set the tone for all that followed. 2. The inaccurate information about CMMI and Agile and the misuse of both resulted in mis- perceptions in both camps about the other. These negative perceptions that position CMMI and Agile at odds with each other arose largely from the following factors: a. Misuse—two decades of experience, first with the Capability Maturity Model (CMM®) and then with CMMI models, in which practices were sometimes misused or applied to (i.e., overlaid on) development activities that may have already been perceived by soft- ware development teams as productive without them b. Lack of Accurate Information—a dearth of accurate information about CMMI in the Agile community and the corresponding dearth of accurate information about Agile me- thods in the CMMI community c. Terminology Difficulties—the use of terminology in CMMI (e.g., discipline, quality assurance, and predictability) and Agile methods (e.g., continuous integration, test- driven development, and collective code ownership) that carries context-specific conno- tations and is thus easily misunderstood and abused d. Top-Down Versus Bottom-Up Improvement Approach—the introduction of an ap- proach that sometimes favors one “voice” (i.e., management versus practitioner) over the other, which neglects the other important voice in how to effectively run the busi- ness
SOFTWARE ENGINEERING INSTITUTE | 1
In this report, we identify and discuss the misperceptions we consider most damaging to a correct understanding of both paradigms, and suggest ways to improve our understanding of both of these powerful approaches for effective software development. In hindsight, we acknowledge that the way in which CMMI was developed and introduced may have helped cause some users to misun- derstand the true message and nature of CMMI1. Such misunderstanding may have led to incon- sistent and ineffective use of CMMI. Further, we identify common misperceptions in the Agile community about CMMI and common misperceptions in the CMMI community about Agile. In many ways, these misperceptions are related to the misuse of CMMI and Agile, but misperceptions can also be attributed to a shortage of accurate information as well as a persistent belief in notions and experiences that are not part of either approach. Some misperceptions of CMMI in the Agile community stem from aspects of the CMM that are no longer found in CMMI. CMMI includes many improvements2 that differentiate it from the CMM, which has not been updated since 19933. Some in the Agile community use CMM con- cepts to judge CMMI unfairly. For example, incorrectly referring to the goal of maturity level 2 as creating repeatable processes persists to this day. In some recent posts to a CMMI-related online forum, a few participants have admitted to not knowing CMMI like they know the CMM. Several participants continue their arguments by writing, “…but if CMMI is anything like CMM, then…” To complicate the problem, there are CMM users who never upgraded to CMMI and therefore persist in using a model that has a less flexible view of software and systems development than the newer CMMI models. There are also former CMM users who upgraded to CMMI but persist in holding on to some of the more dated views from the CMM. An important aside is that when practitioners are naming their activities, the labels “CMMI” and “Agile” are often applied too freely when practitioners are not following either approach properly. These situations contribute to negative perceptions of both approaches. We now know that CMMI and Agile can be used together successfully. Some of the references at the end of this report include experience reports about the successful use of these approaches to- gether. The Software Engineering Institute (SEISM) continues to be interested in the development of Agile methods and in community experiences with both CMMI and Agile.
1 We can’t speak for the Agile community given its diversity; but we feel that to some extent the same could be said for Agile methods. 2 Example improvements include more attention to risk management, integrated teaming, and the work environ- ment and less attention to starting with a fixed set of requirements and doing things “according to a documented procedure.” 3 An incomplete draft of CMM version 2 that included some of these same improvements was created in 1997 but was never formally released. Instead, it served as one of several sources for the initial draft of CMMI. 2 | CMU/SEI-2008-TN-003
2 Origins from Two Extremes The perceived incompatibility of Agile methods and CMMI best practices can, in large part, be attributed to their different origins. In this section, we describe the two extremes from which each approach sprung: Agile from fast-moving markets consisting of small organizations and CMMI from contractually-driven markets consisting of large organizations. These are generalizations, but when you see the details about how each was created, you begin to see why they approach software development from different perspectives. 2.1 THE ORIGINS OF AGILE METHODS The cornerstone of Agile methods originated long before the World Wide Web and collaborative technologies (e.g., wikis and instant messaging). This cornerstone is iterative and incremental de- sign and development (IIDD), a method adopted by engineers over 75 years ago4. Early adopters of IIDD included Department of Defense (DoD) engineers who engaged in propul- sion-related research and development, which included engineering activities tied to hardware not software. An early progenitor of IIDD was Dr. W. Edwards Deming who began promoting Plan- Do-Study-Act (PDSA) as the vital component of empirical engineering. Early adopters of Dem- ing’s teachings in the aerospace industry include NASA (National Aeronautics and Space Admin- istration) and the US Air Force, each of which developed entire systems using time-boxed, itera- tive, and incremental product development cycles. As early as the mid-1950s, IIDD was used in software development resulting in business benefits such as “avoiding management discouragement” and “increasing customer satisfaction.” In fact, a large number of early software development projects, which were often experimental and explora- tive, shared many of the attributes of today’s Agile methods. However, in a systems world dominated by mainframes, COBOL (Common Business-Oriented Language), and the demand to process large and complex datasets, procedural top-down design and development methods dominated and were perceived by many to be the standard. This situa- tion was influenced by the procedural nature of DoD standards and the proliferation of fixed-price contracts5 awarded to suppliers of complex DoD systems (the predominant consumer of computer software at that time). In 1976, Tom Gilb argued that evolutionary development resulted in superior software delivery in his book Software Metrics and launched a movement toward agile, light, and adaptive develop- ment iterations that provided rapid results and more frequently visible business benefits.
4 This short history of Agile methods is provided for the benefit of those who might mistakenly believe that Agile methods are a recent innovation without deep conceptual roots. For a complete history, there are books that describe the events of the past seventy-five years that also contributed to the success of Agile methods. 5 By their nature, the parties to fixed-price contracts assume an unchanging project scope in unvarying develop- ment and use environments. This nature makes it difficult to later modify project direction (without high- ceremony activities) to take advantage of newly-discovered needs and constraints or new technologies.
SOFTWARE ENGINEERING INSTITUTE | 3
As the state of software engineering matured, more formal applications of IIDD became available for example, in Barry Boehm’s 1985 release of The Spiral Model of Software Development and Enhancement. Throughout the 1990s, IIDD gained broad acceptance in the software community in various forms, including rapid prototyping, rapid application development (RAD), and rational unified process (RUP). The seeds of most modern Agile methods were sewn throughout this decade. While most may not expect it, innovative and Agile methods began in the large information tech- nology (IT) shops of several large companies, including an automotive manufacturer and an over- seas bank. XP (eXtreme Programming) began at Chrysler Corporation in 1996 on a project staffed by IIDD advocates Ron Jeffries and Kent Beck, while feature driven development (FDD) started at United Overseas Bank in Singapore, one of Asia’s largest banks. With pair programming and refactoring as XP’s most celebrated features, XP became one of the most recognizable methods of the Agile family. Before the end of the decade, it was clear to many businesses and software engi- neers alike that in many settings, face-to-face communications, rigorous customer interaction, small rapidly moving teams, and frequent delivery of software ultimately produced superior soft- ware. This enlightenment was occurring simultaneously elsewhere and so-called lightweight me- thods proliferated with names such as Scrum, Crystal, FDD, and others. With the proliferation of IIDD methods came the need to coordinate and compare these methods by those interested in their growth and sustainment. The result of this need was a “meeting of the minds” among leaders who were principally responsible for the theory and application of each method. A group of leaders met, including Kent Beck, Ron Jeffries, Alistair Cockburn, Jim Highsmith, Bob Martin, Mike Beedle, Ken Schwaber, Jeff Sutherland, and others who represented the most successful of the new lightweight methods. Modeling their meeting after an earlier meeting of XP enthusiasts in Oregon the year before, these leaders gathered in the Wasatch Mountains of Utah to ski, relax, and ultimately author the Manifesto for Agile Software Development. A subset of the Manifesto authors, together with others like Mary Poppendieck, went on to form the Agile Alliance, a not-for-profit organization dedicated to encouraging the adoption of agile methods. The Agile Alliance primarily focuses on organizing the Agile Conference in the United States every year. While even the organizers of the Utah event expressed skepticism of its outcome, the sessions were a success. The Manifesto documented the guiding principles of Agile development and de- fined a philosophy around a set of existing methodologies. While the first Manifesto for Software Development focused on programming, three years later original Manifesto authors Jim Highsmith and Alistair Cockburn gathered a similar group of early Agile adopters, including Da- vid Anderson, Mike Cohn, Todd Little, and others to establish a set of six management principles known as the Project Management Declaration of Interdependence (DoI) [Anderson 2005b]. The 15 authors of the DoI subsequently formed the Agile Project Leadership Network (APLN), a not- for-profit organization dedicated to encouraging better leadership and management in the IT sec- tor and software engineering profession. 4 | CMU/SEI-2008-TN-003
The Manifesto and Interdependence publications, books written by their original authors, the for- mation of not-for-profit organizations to promote the Agile approach, and the widespread use of the internet for research by software practitioners, has resulted in the rapid growth and broad adoption of Agile methods throughout much of the software engineering profession. Some me- thods, most notably Scrum, continue to grow beyond the software industry into professions that desire the benefits provided by the same basic IIDD concepts first pioneered by Deming and his predecessors. 2.2 THE ORIGINS OF CMMI Before receiving the “CMM” name, the first Capability Maturity Model-like framework was pub- lished in 1989 by Watts Humphrey in his book, Managing the Software Process6. A few years earlier, the U.S. DoD announced a request for proposals (RFP) to address the excessive amount of money being spent on software that was either never delivered or delivered late with little of its expected functionality. The contract awarded on the basis of the RFP was to establish what we know today as the Soft- ware Engineering Institute (SEI) at Carnegie Mellon® University. The SEI brought together repre- sentatives from academia, research, and industry to expose and promote practices demonstrated to be successful at avoiding the failures so beleaguering DoD software acquisition efforts. Carnegie Mellon’s framework of practices to address the DoD’s issue became the CMM. If we look at the genesis of the CMM, it predates the internet and nearly everything associated with internet technology. For that matter, CMM predates many software development, deploy- ment, and infrastructure technologies, languages, and methods. We’ve all learned a lot in the past 20 years. When the DoD set out to address their “software dilemma,” the software world was dif- ferent than it is today. To color this context even further, everyone working to develop the initial CMM was looking for the solution to the “software problem” from the perspective that software is a component of a larger system and that if it failed, lives would be lost (e.g., aircraft, ships, weaponry, medical de- vices). Systems were evolved using careful and deliberate development paths according to lower- risk, standardization-heavy and contractually-driven relationships between the developer and the customer. In today’s frequent discussions of increasing globalization and the important role played by trust (i.e., level of social capital7) [Fukuyama 1995] in making effective collaboration happen across stakeholders, one might describe such a development context as exhibiting low trust. Users were typically not direct contributors to the evolution of the end product prior to field testing. They instead had to depend on the contracting relationship, requirements, and standards to deliver the product they needed.
6 This short history of CMMI focuses more on the past twenty years, but as in the case with Agile, the roots for many of the product development, project management, and process concepts found in CMMI have a long his- tory. 7 This definition of trust may be clearer but more elaborate: “the willingness of a party to be vulnerable to the actions of another party based on the expectation that the other will perform a particular action important to the trustor, irrespective of the ability to monitor or control that other party” [Mayer 1995]. For the purposes of this report, the particular question being asked is whether there is a level of trust between the project and its cus- tomer that will allow them to effectively negotiate scope as the project progresses, without the customer requir- ing detailed accounts of project effort.
SOFTWARE ENGINEERING INSTITUTE | 5
These comments may be an over-generalization, but they are intended to summarize the DoD software acquisition environment that existed at the time. Further, these comments explain why the practices in CMMI sometimes exhibit some of these same high ceremony and low trust cha- racteristics found in the high-risk, government-contractor environment in which software failure could equal lives lost. Also, within the high-risk government-contractor environment at that time, the prevailing pattern of infrequent and monolithic deliveries contributed to the high costs associated with deployment, upgrade, and replacement (e.g., software embedded in fighter aircraft in the 1980s could not be upgraded over the air or over the internet). Hence, getting it right the first time was critically im- portant. Furthermore, most of the organizations involved in this contractual environment were large organizations working on large complex projects. Finally, the use of public money in government contracting (or similar high-visibility and high- stakes situations) requires a level of accountability by all those involved that often drives all par- ties toward risk-averse behavior bordering more on protecting one’s own interests8 than on find- ing the most efficient win-win solution. Ceremonial but perfunctory activities help address the often competing and incompatible self-interests of all parties, but make operating in an open and transparent manner challenging, and reinforce the perception of low trust9. Within a few short years, the CMM was expanded into several other models; these were point solutions developed to address non-software development projects. Also, the CMM and these va- riants increasingly became used internationally and by commercial industry. Organizations at- tempting to adopt more than one model on any given project quickly realized the challenges of doing so and petitioned the SEI to consolidate the various CMMs into one model, which in 1998 led to a joint industry, government, and SEI team to create CMMI. Of course, over the years that the CMM and CMMI have been maintained, inputs on what consti- tutes good management and development practice have increasingly come from a wider variety of industries and from users around the world. As a result, new ideas, standards, and practices are continually being incorporated into what is now the CMMI Framework10.
8 Two somewhat extreme examples of protecting one’s own interests include (1) a contractor that under bids, hiding the real cost of the work to win the contract, then treating every change as an opportunity to reclaim some of the investment; or, conversely (2) a government program manager who strongly encourages “cutting corners” to meet aggressive cost targets, only to leave the resulting disaster-in-wait to fall on “the watch” of his or her replacement. 9 When taken to an extreme, ceremonial dependence on plans, processes, and standards can replace common sense, usurp accountability, and adherence to them can be used as an excuse for poor project performance and poor product functionality. 10 In August 2006, CMMI for Development, Version 1.2 was released and in November 2007, CMMI for Acquisi- tion was released. A third model covering service delivery is under development. Most of the comments in this report apply to the first of these, CMMI for Development, as there has been less exposure to the newer models in the CMMI family. 6 | CMU/SEI-2008-TN-003
3 Factors that Affect Perception Perceptions that position CMMI and Agile at odds with each other largely arise from factors present in the relatively more volatile commercial software development community. These fac- tors may be generally attributed to market forces and human nature. We’ve described them as (1) misuse, (2) lack of accurate information, and (3) terminology difficulties. 3.1 MISUSE Whether analyzing the CMM or CMMI, there is one thing shared by both works that makes them unique—they are models, not standards, for improving product quality and process performance. However, for nearly two decades, the software industry has experienced the result of people mi- susing appraisal ratings as entry criteria, confusing appraisal ratings for measures of business per- formance, and misapplying a model as a standard in an environment in which products are created to meet contractual requirements. Used in this way, CMM and CMMI best practices were misinterpreted and misused. It is not an exaggeration to say that any approach to improve an organization’s achievement of business ob- jectives in such an environment would have difficulty overcoming an emphasis on RFP require- ments and keen competition for multi-year contracts. Does this situation mean that CMMI is wrong for software? Not in the least. It simply illuminates the following reasons why some perceive CMMI to be incompatible with Agile ideals:
The context from which the CMM and CMMI originated was specific to a particular cus- tomer base having unique challenges and characteristics of high risk and low trust.
The CMM and CMMI were a new paradigm introduced into a large (and dominant) industry where paradigms, including the attitudes and beliefs associated with them, were in place for many years (e.g., command and control).
Agile ideals developed as a backlash against the inefficient software development patterns that arose in this industry. These points describe the context in which the CMM and CMMI were developed. This context enables us to understand some of the characteristics of the CMM and CMMI and how they have been used over the past two decades. While the language of CMMI, admittedly, may retain some of the flavor and phrasing of this context, each release of a CMMI model grows further away from these roots to embrace a richer and more dynamic set of contexts11.
11 Users of older model versions may not fully make the transition as new versions are released, and older beliefs and values may persist. Further, while CMMI and SCAMPI materials continue to evolve, it is simply not possible to bring all users rapidly forward to the newer versions, government edicts and SEI encouragement notwith- standing.
SOFTWARE ENGINEERING INSTITUTE | 7
Further, this context is not the sole element determining how CMMI should be used, when and where it can be used, or what defines whether CMMI is being properly used. Nor for that matter, does this context determine its applicability in other contexts. The challenge is for the broader community to identify the practices and methods (or practice implementations) that enable orga- nizational maturity in more dynamic contexts (e.g., internet commerce, social networking, and games development). An increasing subset of both CMMI and Agile method users are trying to do just that. 3.2 LACK OF ACCURATE INFORMATION As we write this report, the availability of accurate information has improved. However, these improvements are new, thus, the pre-improved situation is still the current one for many. Internet searches for information (papers, presentations, trade articles, blog posts, discussion group posts, and wikis) on Agile and CMMI sorted by year yields telling results. It’s no surprise that in many of the writings of the original Agile Alliance (those who signed the famous Agile Manifesto and Agile Principles), much of the impetus for their collective philosophy and methods were a direct counter-point to what they (often justifiably) perceived as heavy- handed, wasteful, over-burdened and ceremonial processes. “Oil and water” is not an uncommon expression used to refer to the relationship between Agile methods and CMM/CMMI in the writ- ings and postings of Agile supporters. If we account for the historical differences and challenges described above, this is no surprise. What is interesting is that it is easier to find far more material on CMMI from the Agile perspec- tive than on Agile methods from the CMMI perspective [Levine 2005, SEI 2008e]. This observa- tion leads many to conclude that Agile methods were largely ignored by CMMI users, which is not far from the truth. Although two noteworthy articles were written from this perspective in 2001 (one by Hillel Glazer in CrossTalk – The Journal of Defense Software Engineering, and one by Mark Paulk [an author of the CMM], which was broadly distributed by the IEEE Computer Society Dynabook), with few exceptions, these perspectives were received with open skepticism from both camps, but otherwise largely ignored. Discussion of Agile methods in CMM/CMMI settings was frequently peppered with anecdotes equating Agile with “no discipline.” In fairness, the XP community (unintentionally) invited this censure. Even the name “extreme programming” conjures up an image of skateboard parks or the rule-breaking anarchy of off-piste snowboarding. Certain readings of the Agile Manifesto (ex- cerpted in Section 5) might bear resemblance to this perception. With Agile proponents being per- ceived as openly “hostile” to CMM/CMMI, and CMM/CMMI being too often misapplied as de- scribed above, the few attempts to draw attention to the possibilities of a peaceful co-existence between CMMI and Agile were easily lost in the noise of other competing messages. In searching conferences, little material on Agile and CMMI used together exists prior to 2005. In late 2005, the NDIA (National Defense Industrial Association) introduced an Agile and Lean track to their CMMI Technology Conference and User Group event, and in 2006, the SEI added a similar track to the SEPGSM 2006 conference. In 2007, both events grew their respective Agile tracks, while similar tracks appeared at events of other process-oriented organizations. Since these events, material placing Agile and CMMI used together in a positive light began to grow, and at the most recent Agile conference (Agile 2007), a presentation was given on a successful introduc- tion of Scrum in an organization that had achieved maturity level 5 [Sutherland 2007]. 8 | CMU/SEI-2008-TN-003
Agile and CMMI have been available simultaneously since 2001, and while participation at con- ference sessions has continued to grow in subject areas combining them, not before 2005 was there a major presentation on the successful merging of the two concepts12 [Anderson 2005a]. It is also worth noting that the Agile and CMMI communities don’t mix much. Few attendees of Agile conferences also attend CMMI conferences and vice versa. Even more surprising, rarely do the thought leaders in either community publish in the same sources. Nonetheless, continuing research on the topic reveals that although momentum seems to be gain- ing on the side of the successful merging of Agile and CMMI, it has yet to overtake (let alone ex- tinguish) the existing perception that the two ideas are incompatible. It may be worth addressing the demographics in the software development industry. How much of the industry was Space and Defense in 1985? How much of it is Space and Defense today (meas- ured by participation of the 15+ million software engineers worldwide)? Of those not in Space and Defense, how many are so young that they never used a third generation language, such as C, or worked on a true waterfall-style project? As a result of these demographics, a growing number of members in our profession demonize traditional methods (which for some, CMMI symbolizes) based on superstition or hearsay rather than experience. We’ll return to the topic of available information in section 6.3, Current Published Research. 3.3 TERMINOLOGY DIFFICULTIES CMMI, Agile development, traditional development, and product development in general all have their own vocabularies. However, the overlap in these vocabularies includes terms that are given different or even niche meanings. More important is when terms introduced in one vocabulary have a history of use elsewhere and the term becomes associated with its previously-used context. For example, technical data package (TDP) is a term used in CMMI that is intended to refer to the collection of product-oriented documentation associated with a product being developed that describes the product from a technical standpoint. It is typically used for further development, operation, installation, training, repair, support, troubleshooting, or maintenance of the product. The TDP can include text, diagrams, drawings, designs, specifications, and/or data sets. (In a small Agile project, the software code itself may largely suffice.) However, the term TDP has a meaning in systems acquisition where it refers to a specific deliverable that includes specific doc- uments. Not all projects create that specific set of documents, nor is it reasonable for them to do so. Many people unfamiliar with the holistic purpose of CMMI can easily misapply their previously- learned systems acquisition definition of TDP and thus assume that CMMI requires, for example, a MIL-STD-498 version of a TDP. This assumption is not the case. In actuality, CMMI uses the term TDP far more generically—an interpretation more likely to match what most small or Agile projects actually do and, possibly, what most successful projects do whether or not they are under contract with a government agency.
12 At the Agile conference in Denver in 2005, David Anderson recounted his experience developing Microsoft Solutions Framework for CMMI Process Improvement as an Agile method compatible with CMMI maturity level 3.
SOFTWARE ENGINEERING INSTITUTE | 9
Also, the word predictable is a locus of the confusion and misunderstanding between the tradi- tional and Agile development communities. Many in the Agile community believe that software projects cannot be predicted with any great precision. These believers say, “Perfect is the enemy of good enough, so don’t try to predict, just react and re-factor.” However, CMMI uses the word predictable in a more subtle sense13. Predictability does not arise from a detailed project plan up front that covers the entire project lifecycle (nor does CMMI require or generally expect this type of predictability to be possible). The Agile community might find a lot of benefit and interesting results from experimenting with this higher-level concept of predictability. For example, David Anderson has achieved predictabil- ity in sustaining engineering at Corbis with a release every two weeks, but the content of the re- lease is not bound until five days prior to release due to the unpredictable nature and dynamic cha- racter of the work scope [Anderson 2008]. Through better process management and improvement, he’s been able to meet CMMI rigor for predictability while simultaneously using an Agile ap- proach that adapts to the unpredictability of the work and the market.
13 In CMMI at maturity level 4, predictions are derived from an understanding of variation at the process-step level combined with probabilistic or statistical models that predict the ranges in quality and process performance that can be expected as project final or interim outcomes. For example, by investigating and characterizing the de- fect injection and removal behavior of individual design, coding, and verification subprocesses as a function of team skill and product complexity, and statistically modeling the relationship among these attributes and project final or interim outcomes, a project can periodically evaluate whether it is still on track to achieving its quality and process performance objectives. In a similar manner but by considering team velocity and the attributes af- fecting team velocity, an Agile project can be “predictable,” even though the scope, schedule, and budget of the project are changing and reacting to feedback and market conditions. 10 | CMU/SEI-2008-TN-003
4 The Truth About CMMI In CMMI, process management is the central theme. It represents learning and honesty as demon- strated through work according to a process. Process also enables transparency by communicating how work should be done. Such transparency is within the project, among projects, and being clear about expectations14. Also, measurement is part of process and product management and provides the information needed to make decisions that guide product development. However, if CMMI is used in the pursuit of maturity level numbers, the resulting process im- provement efforts can sometimes lose sight of customer value, product, project value, and practic- al business goals. Also, when CMMI users establish (sets of) standard processes for use by the organization and its projects, they sometimes fail to ensure that these processes are (1) successfully deployed to all new projects; (2) periodically revised based on lessons learned from use; (3) compatible with in- dividual and team practices and flexible enough for teams to adapt the processes to their needs according to their experience; and (4) written using language and formalisms that practitioners understand. The result can be rejection of the organization’s process improvement efforts by prac- titioners. There is a balance to achieve across organizational, project, and individual prerogatives and re- sponsibilities that are a function of risk, task maturity, trust, and other business, fiduciary, and cultural considerations. If the balance is weighted too heavily in favor of the organization, projects and practitioners may lack the flexibility they need to be successful and motivation fails. On the other hand, too much flexibility can expose the organization to excessive risk (e.g., from mis-aligned teams) and missed opportunities for organizational learning that over the longer term might lead to improved product quality and productivity. It is difficult to achieve the right bal- ance. 4.1 CMMI IS A MODEL, NOT A PROCESS STANDARD A plausible rationale for many of the challenges faced by organizations using CMMI is likely the fact that most processes implemented using CMMI (as witnessed by the authors and garnered from feedback from others who use CMMI) fail to distinguish that CMMI is fundamentally a model. Instead of working with CMMI as a model, they work with CMMI as a standard. A stan- dard is an auditable, testable, compliable work with a narrow field of distinct, acceptable, and demonstrable outputs with little variation from one implementation to the next. This view of CMMI as a standard is a complete misuse of a model. To reiterate from the model itself, “CMMI contains neither processes nor procedures;” the lists of typical work products, for example, are examples of process outputs. They are not exhaustive lists of required process out- puts nor are they mutually exclusive of other possible process outputs. The practices are not steps in an organization’s set of standard processes, and they are not activities that necessarily occur neatly within a specific business process.
14 In reality, CMMI and Agile share many of the same values.
SOFTWARE ENGINEERING INSTITUTE | 11
They are process-oriented activities that when used collectively can attain the process area and business goals to improve the real-life business activities of an organization, wherever and when- ever those activities may exist. CMMI’s practices are also meant to encourage an organization to experiment with and utilize other models, frameworks, practices, and process improvement ap- proaches, as appropriate, based on the organization’s needs and priorities. In some important ways, CMMI is an “ivory tower” of theoretical concepts born of decades of research and practical application. It is a collection of the activities to be expected of an organiza- tion as it sets out to improve its processes. CMMI is not and never was meant to be a replacement or a definition of anything in the real world. That is what a model is. It is an ideal from which we are meant to learn and relate to actual situations. All too often, CMMI has been applied rather than implemented. The standards-centric application of CMMI has contributed to some spectacular failures and losses of time and money. The key difference between applying and implementing CMMI is that applying usually appears as a super- imposition (or overlay) of model practices onto existing activities with an expectation of produc- ing the example work products found in the model, rather than seeking the natural products of the organization’s processes. This misplaced focus is often the by-product of an overly strong focus on achieving a particular appraisal rating. In contrast, implementing CMMI is using the model in the same way that engineers and architects use models: as a learning tool, a communication tool, and a means of organizing thoughts. The more implementation-oriented an organization is, the more improvement-centric it is; thus, the focus is on maturing and growing process capabilities rather than ratings. The ratings are the by- product of the improvements in these implementations; and the model serves as a tool to holisti- cally examine the organization’s processes and performance and determine possible areas for im- provement. When viewed holistically, CMMI’s ultimate goal (i.e., continuous process improvement) is to cause an organization to become less wasteful, leaner, and more in touch with actual development progress. Ultimately, both Agile and CMMI, especially in high-trust environments, expect organi- zations to see gains in productivity by eliminating unnecessary effort. It’s true that implementing Agile methods will often eliminate many unproductive efforts and behaviors at the project level. However, even with Agile retrospectives, what CMMI offers beyond Agile is an infrastructure of organizational learning and improvement that benefits the projects even before they begin. When using CMMI as a model for process improvement rather than a standard for it, organiza- tions are free to implement whatever applications of process improvement suit their needs and they can see organization-wide progress beyond the project level and beyond immediate projects and developers. CMMI provides an approach to improvement in which processes and improve- ments to them are supported by the organization, persist beyond their original implementation date, and do not encumber development teams with waste (e.g., from re-inventing a process al- ready known to work well in similar circumstances). 12 | CMU/SEI-2008-TN-003
4.2 PROCESS AREAS, NOT PROCESSES CMMI is comprised of goals and practices organized into process areas. To simplify this discus- sion, no distinction is made between the types of goals or practices—suffice to say that each goal is composed of several practices. An example goal (SG 1) and practice (SP 1.2) from CMMI for Development, Version 1.2 reads as follows: SG 1 Establish Baselines Baselines of identified work products are established.
SP 1.2 Establish a Configuration Management System Establish and maintain a configuration management and change management system for controlling work products. Process areas are not processes themselves. The processes themselves could be operating wherev- er and whenever and in whatever sequence necessary to perform the work of the business. Satisfy- ing (i.e., achieving) goals implies that those areas of activities where processes are happening are being improved to some degree. At this point, you easily can see that misinterpreting CMMI’s process areas as actual processes can cause an enormously misleading effort and thus waste and abuse. Waste is found when organ- izations apply practices not intended to be performed as written, and abuse is found when evi- dence is not accepted in an appraisal because it lacks of conformity with the process areas, which are considered defined processes. Goals are the only required components of CMMI. The practices within the goals are expected, but not required. What this statement means is that to satisfy a goal, some form of activity must take place that generates a state, output, or persistent condition in which a goal is satisfied. In the absence of any such activity, the practices themselves suggest what might be a suitable approach to satisfying a goal. Again, these are not necessarily practices to be incorporated as is within a given organization’s processes, but instead are recommended starting points for the purpose of improving an organization’s satisfaction of the goals of a process area. Performing these practices in the absence of a business value for doing so is as unlikely to im- prove an organization’s processes as not performing the practices at all. What is lacking in rote practice performance is the understanding of the rationale or motivation for the practice and the potential value the practice has toward achieving a business objective important to the organiza- tion. Although CMMI allows the satisfaction of alternative practices that meet the goal but do not ex- plicitly implement the practices, this concept is poorly understood by users. Further, there is a fear among users that the use of alternative practices will be viewed suspiciously in an appraisal.
SOFTWARE ENGINEERING INSTITUTE | 13
Within the practices there is a tremendous amount of informative material. Including this infor- mative material is the only way to describe a model15. Narrative is necessary to understand, interp- ret, and implement the model’s practices and goals. On the plus side, informative material helps to explain the purpose of the practice and provide considerations relevant to how it might be imple- mented. However, for people and organizations that do not have the skills or experience to understand, interpret, and implement the practices and goals to their specific contexts, the informative material takes on the mantle of requirements. This situation is also true for appraisal teams and lead ap- praisers. Thus, the over-emphasis on specific model examples, typical work products, and sub- practices often results in prescriptive processes with little gain in quality or organizational per- formance. This concept-based disconnect is compounded by an approach toward CMMI use that can be apt- ly described as pathological box-checking and an over-zealous contracting or “small-picture” au- diting attitude that seems willing to sacrifice what is good for people, projects, customers, prod- ucts, and the long-term health of each in favor of supporting the generation of appraisal evidence. 4.3 SCAMPI APPRAISALS With the proper understanding that CMMI is a model, the SCAMPISM (Standard CMMI Appraisal Method for Process Improvement) appraisal method is designed to determine whether process improvement (i.e., process maturation) is occurring as a reflection of model (e.g., CMMI model) practices having been implemented. Even the term used for CMMI artifacts points in the direction of practice implementation rather than practice application. SCAMPI appraisals are not audits or tests. Instead, a SCAMPI appraisal is a process story in which the elements of the story include the following:
the CMMI model
the organization’s development practices
acculturation of those development practices
the improvement activities within the organization’s broader set of practices
the ability of project participants to articulate their development practices
15 Another way to express the need for informative material is that the value in a CMMI practice does not lie only or even principally in the statement itself. Rather, the statement is a summary of a larger concept comprised of activities or operations that will achieve a particular process area goal. To quote from the SEI website (i.e., http://www.sei.cmu.edu/cmmi/adoption/cmmi-informative.html), “[I]nformative material supports correct interpre- tation and implementation of CMMI practices, but is neither required nor expected, nor is it to be used as part of a checklist in an appraisal.” The informative material “takes on a critical role in explaining …the intent of the practice, [and]… the practice’s dependencies with other parts of the model.” If the intent is met in some other way or if the goal can be achieved through activities different from those specified in the CMMI practice state- ments, then the CMMI model material associated with that practice or goal need not be further pursued from a maturity rating standpoint. (Of course, it is possible that further process improvement is possible through further consideration of informative material, but such pursuance must be weighed against higher priorities and better opportunities for improvement to the business.) 14 | CMU/SEI-2008-TN-003
It is the job of an appraisal team, led by a lead appraiser, to discern the improvement activities of the organization and to determine whether the practice implementation indicators16 objectively indicate whether CMMI was implemented. It is how the above process story elements hold to- gether that determine the outcome (often a rating) of the SCAMPI appraisal. If an organization has been engaged in an effort to improve its processes, Agile or otherwise, the SCAMPI method is used to ascertain which process area goals are being met. It then becomes incumbent on the appraisal team to understand the context of the organization’s practices and the relevance of their activities toward process improvement. In many cases, these activities—based on the organization’s context—may not appear in a form or format familiar to those whose expe- riences are limited to non-Agile development approaches. Nonetheless, it is the appraisal team’s responsibility to interpret CMMI model practices and the organization’s artifacts of process im- provement so that the appraisal team can correctly determine the overall process maturity and/or capability profile of the organization being appraised. Agile methods are not addressed specifically in either the model or the SCAMPI method because Agile is a particular approach to development activities that contrasts with other approaches to development. CMMI, not being a development approach, and SCAMPI, being agnostic on the subject of development approaches, are not irrelevant or of no value because an organization chooses to use Agile principles. An organization’s development approach is its choice.17
16 The fundamental idea of practice implementation indicators (PIIs) is that the conduct of an activity or the imple- mentation of a practice results in “footprints” (i.e., evidence that provides a basis for verification of the activity or practice implementation). 17 Likewise, the SEI has been asked over the years why CMMI does not promote specific methods, tools, or tech- niques such as function points, RUP, or Six Sigma (or you name it). Again, by being a model that is imple- mented as opposed to a standard process that is applied, CMMI, by its nature, must remain agnostic to such selections, leaving the organization free to select the methods, techniques, and tools, most appropriate for its use.
SOFTWARE ENGINEERING INSTITUTE | 15
5 The Truth About Agile The Agile Manifesto reads as follows: ―We are uncovering better ways of developing software by doing it and helping others do it. Through this work we have come to value [the following]: individuals and interactions over processes and tools working software over comprehensive documentation customer collaboration over contract negotiation responding to change over following a plan That is, while there is value in the items on the right, we value the items on the left more.‖ The Agile Manifesto is frequently read in such a way that ignores the last line (by too many Agile proponents and detractors alike) and where the “things on the right” (items commonly found in too many plan-driven, contractually-driven, standards, and audit-driven environments) are not merely valued less, they effectively have zero value. The Agile Manifesto has been frequently cited by Agile proponents as justification for not having processes, for not documenting their work, and for not having plans. This interpretation gives jus- tifiable cause to Agile detractors to accuse proponents of being “undisciplined lackeys.” Similar- ly, Agile detractors abuse the Agile Manifesto by using the same exact mechanism. By attributing a value of zero to the “things on the right,” detractors assume the worst of Agile proponents. However, these interpretations of the Agile Manifesto are incomplete and disingenuous. At a genuine level of discourse, who wouldn’t value the things on the left more than the things on the right? Keep in mind the paradigm from which the Agile Manifesto is trying to depart. Even without researching the discussion among the signers, the source of the Agile Manifesto should not be difficult to guess. In too many implementations there were plans, processes, and standards that were overly detailed, rigid, and just plain abused to the detriment of people, projects, prod- ucts, customers, and technology. Common practices associated with Agile approaches include the following:
Iterative and short incremental time-boxed development is used.
There is frequent and ongoing customer feedback. Customers are commonly embedded with developers. In fact, extensive use of tacit knowledge is a key in Agile development and prac- tices. Use of tacit knowledge is of most benefit in a high-trust environment that can directly benefit from the low transaction costs of software engineering activities. The risks from the use of tacit knowledge, though often not explicitly managed, are mitigated by the following nature of the domain:
the length of the project (and the expected lifecycle of the product)18
the self-documenting nature of modern code
18 Agile teams and projects often assume low staff turnover (rightly or wrongly) because of the short nature of projects and the small size of teams. 16 | CMU/SEI-2008-TN-003
the use of tools that can reverse engineer designs19 and other artifacts as well as conti- nuously integrating code
the small size of the team.
Each increment potentially delivers value (e.g. working code). Value can also be what not to do.
The delivered product is fully tested for each delivery. Products are often created by writing rigorous tests first, then evolving the product to pass the tests.
Everyone owns quality, including the customer (or customer representative). Development teams are often multi-disciplined, in which all developers are generalists and collectively “own” the product. Customers are equally accountable to ensure the product meets their ex- pectations. Products are prevented from advancing until a common understanding of the product’s functionality, inclusions, and exclusions are established. There is a fail early atti- tude with an underlying assumption that the cost of failure is low to negligible.
Detailed requirements are developed just in time and evolve along with the understood prod- uct. Effort is not invested (i.e., wasted) by speculating on requirements or expectations. As the product to be delivered evolves, the requirements become more and more detailed20.
Change is expected, embraced, welcomed, and/or accepted at any time. Acknowledged sources of changes in traditional (i.e., serial) development are changes to requirements that were baselined too early (from an Agile or Lean perspective). Furthermore, the expectation of changes drives design decisions. By design, the product and architectural components are not locked in so that developers can perform continuous trade-offs until all the component requirements are available21.
Self-directed, empowered teams are used. Task-mature, well-trained, cohesive, high-trust, small teams who take ownership of the product need little direction, management, or task- feeding. Broad product, project, and business goals are shared with the team so that informed managerial, organizational, and technical decisions can be made by the team to determine what must be done, by when, by whom, and to what standards. Teams are allowed to re- spond to current conditions rather than adhering to over-emphasized, over-detailed plans. Thus, plans and planning are actually updated frequently and communication with superiors (whether organizational or team) often takes place daily or nearly daily. Highly ceremonial
19 Again, there is a bias against documentation. People assume that documentation is complete and up to date, which is often not the case. Without sustained effort over time to keep it up-to-date, it can become misinforma- tion. Documentation is often a poor substitute for communication, especially for a co-located team. 20 There is a recent movement toward articulating this just-in-time requirements method through the use of Real Options Theory. The key proponents are Chris Matts (from London), Olav Maassen (from Amsterdam), and Kent McDonald (of the Agile Project Leadership Network, from Iowa). 21 Traditional assembly line production of code assumes a cost of change curve that accelerates with late discov- ery. Hence, early discovery through analysis methods and inspections is desirable [Boehm 1976]. Some Agile proponents (e.g., for eXtreme Programming) have argued that the cost of change rises quickly and asymptoti- cally levels out [Beck 2000]. Actually, they are both often wrong assumptions—cost of change is contextual and depends on where the capacity-constrained resources lie in the process flow [Anderson 2003]. Cockburn showed that slack resources in development can be used to fix defects at no cost to productivity (i.e., through- put) and hence, in some situations the cost of change in coding is as Beck suggested negligible [Cockburn 2006].
SOFTWARE ENGINEERING INSTITUTE | 17
status meetings with management or reviews with customers are replaced with frequent inte- ractions with these stakeholders.
The process is periodically evaluated and adjusted. Task-mature teams take ownership of the process they employ and adjust it based on feedback on what works and what does not. In general, attributes of successful Agile implementations include the following22:
small teams consisting of approximately ten people
an involved customer
“rolling wave” or continual planning
co-located and cross-functional teams
organizations that are not in the habit of breaking up teams until each member is individually competent at Agile Agile is not successful when the following is true:
lack of processes
lack of discipline
absent a role for plans or planning The Agile Manifesto has a list of general principles attached to it. ―We follow these principles: Our highest priority is to satisfy the customer through early and continuous delivery of valuable software. We welcome changing requirements, even late in development. Agile processes har- ness change for the customer's competitive advantage. We deliver working software frequently, from a couple of weeks to a couple of months, with a preference to the shorter timescale. Business people and developers must work together daily throughout the project. We build projects around motivated individuals. Give them the environment and support they need and trust them to get the job done. The most efficient and effective method of conveying information to and within a de- velopment team is face-to-face conversation. Working software is the primary measure of progress. Agile processes promote sustainable development. The sponsors, developers, and users should be able to maintain a constant pace indefinitely. Continuous attention to technical excellence and good design enhances agility. Simplicity—the art of maximizing the amount of work not done—is essential. The best architectures, requirements, and designs emerge from self-organizing teams.
22 Some organizations, which are adopting both CMMI and Agile methods, characterize the situations that are more suitable for the implementation of Agile methods according to the level of trust, customer/end user availa- bility, project scope, scale, expected lifespan of the product, cost of delay, value of early (partial) delivery, and cost of failure present in the situation. 18 | CMU/SEI-2008-TN-003
At regular intervals, the team reflects on how to become more effective, then tunes and adjusts its behavior accordingly.‖ Each word in the Agile Manifesto and the guiding principles was attained via consensus—a process familiar to anyone having been on a SCAMPI appraisal. Each word carries meaning. Note the use of priority, most, primary, sustainable, excellence, and continuous. These are conditional and qualifying words, not absolutes. Anyone who interprets these words as absolutes are misusing Agile ideas just as they would if they were to interpret of CMMI’s typical work products as abso- lutes.
SOFTWARE ENGINEERING INSTITUTE | 19
6 There Is Value in Both Paradigms CMMI and Agile are compatible. At the project level, CMMI focuses at a high level of abstrac- tion on what projects do, not on what development methodology is used, while Agile methods focus on how projects develop products. Therefore, CMMI and Agile methods can co-exist23. There can be much value gained from Agile and CMMI synergies. Today, many CMMI-adopting organizations have Agile development teams. Conversely, CMMI can be effectively introduced in an Agile setting where an iterative, time-boxed approach is used, which is perfectly compatible with CMMI. CMMI and Agile can complement each other by creating synergies that benefit the organization using them. Agile methods provide software development how-to’s that are missing from CMMI best practices that work well—especially with small, co-located project teams. CMMI provides the systems engineering practices that help enable an Agile approach on large projects. CMMI also provides the process management and support practices that help deploy, sustain, and conti- nuously improve the deployment of an Agile approach in any organization. 6.1 CHALLENGES WHEN USING AGILE The major challenge when using an Agile approach24 in a large project is keeping the small teams aligned and coordinated for the duration of the project to ensure its success while adhering to Agile’s team-focused principles and values. Maintaining alignment and coordination across a dis- tributed project requires that someone (possibly a team) or a mechanism maintain coherence (i.e., unity, logic, and consistency) of the following:
overall system capabilities to be developed, including non-technical requirements
scope, quality, schedule, cost, and risk tradeoffs
product (or service) architecture If coherence is not achieved, alignment becomes divergent on multiple fronts. For example, one common Agile practice, re-factoring, often doesn't scale up well when multiple development teams must cooperate on the same product. Through a combination of activities such as release planning, test planning, continuous integration, and employing feature-oriented rather than com- ponent-oriented teams, Agile approaches have been scaled to work in larger projects. From a systems engineering perspective, these activities are challenging for large, complex projects to perform well:
establishing overall product objectives and a project vision
managing the allocation of requirements to teams
23 The references at the end of this report include publications and presentations that also characterize CMMI and Agile as complementary and synergistic [Anderson 2005, Boehm 2003, Canditt 2008, Santos 2007, and Suther- land 2007]. 24 Of course, this is a challenge common to all large, complex projects. 20 | CMU/SEI-2008-TN-003
defining and maintaining interfaces and constraints among teams (for both the product and the process)
maintaining effective product integration, verification, and validation strategies for the over- all product (or service)
coordinating risk management across the project These alignment and coordination activities, necessary in larger, complex projects, are described in the systems engineering practices found in the Engineering, Risk Management, and Integrated Project Management process areas of CMMI. Thus, CMMI provides a “safety net” for large projects that helps reduce the risk of something going very wrong. Some Agile proponents acknowledge that many Agile methods have yet to be proven effective or possible as the core product development process on large projects25. However, there has also been increasing success scaling Agile approaches for use on large projects by introducing a top layer of coordination (e.g., a Scrum of Scrums) and by providing explicit and early attention to non-functional requirements and product architecture26. Alternatively, Agile methods are also in use on features or components of larger systems in which the Agile team is, in effect, isolated from the rest of the product development team. The software development community is increa- singly learning how current Agile methods can be effectively extended to work on large-scale, complex projects. Agile methods generally lack practices and guidance for implementing and supporting an Agile approach across the organization. While some larger companies (e.g., CapitalOne, BMC Soft- ware, and Yahoo!) are pursuing large-scale enterprise adoption of Agile, the level of documenta- tion and implementation guidance available to support such adoption is relatively modest27. Agile implementations will not “stick” or improve without an organizational context that supports process definition, measurement, feedback, training, and improvement—mechanisms that are de- scribed in CMMI. CMMI practices also address institutionalizing and improving deployed processes, be they Agile or not; therefore, CMMI supports a long-term view of process adaptation and improvement across the enterprise as a whole. Of course, many organizations have identified and implemented project management, software development, and process improvement practices without the benefit of CMMI. The book Micro- soft Secrets discusses Microsoft’s seven-year journey to improve its product quality [Cusumano 1995].
25 A lot depends on the definition of large and which Agile method is being used. The first FDD project (1997- 1999) was characterized by 50 people, 18 months, and 1.5 million lines of Java. Two years later, Jeff DeLuca led a 250 person project in New Zealand. Hence, there is relatively early literature and evidence that Agile tech- niques can scale up [Highsmith 2002]. 26 A continuing challenge experienced in many Agile implementations is ensuring adequate and early attention to non-functional requirements and product architecture [Cao 2008]. Practices that address this challenge are found in the Engineering process areas of CMMI. Relative to its own Agile approach, the SEI is currently colla- borating with NAVAIR on extending the Team Software ProcessSM (TSPSM) to address systems engineering and is also experiencing promising results from doing so [Carleton 2006, SEI 2008d]. 27 A notable exception is [Schwaber 2006], which provides guidance and case studies on how to introduce SCRUM to an enterprise; but may have applicability to other Agile approaches and to large projects as well.
SOFTWARE ENGINEERING INSTITUTE | 21
Arguably, in some of its divisions Microsoft put in place many of the practices required to achieve maturity level 3 in a manner reflecting the challenges of their environment. Therefore, given enough time, organizations can address what is missing from their business and implement a solu- tion without the benefit of a model such as CMMI. However, when used with care, the appropri- ate model can certainly make the journey easier. Impediments to introducing an Agile approach also include factors common to the introduction of any new technology: lack of management support and general resistance to change. CMMI and the process improvement experience that has grown around it can address these issues. For exam- ple, with the introduction of an Agile method, management often fears a loss of control. Blending a top-down approach (i.e., CMMI) with a bottom-up approach (i.e., Agile) may go a long way in helping to reassure management that its concerns are being addressed in the improvement effort. 6.2 CHALLENGES WHEN USING CMMI There are challenges to successfully using CMMI best practices as well. No development ap- proach or methodology can effectively address all difficult challenges or situations [Elm 2007]. Just because an organization has been appraised at a particular CMMI maturity level is no guaran- tee that a particular project in the organization will succeed. However, organizations using CMMI can fail (and some are failing!) because they misuse the model or pursue process improvement and subsequent appraisal with misguided motivation or with imprudent leadership. Sometimes, in the rush to achieve a maturity level, the focus on improving organizational perfor- mance is lost. The wrong standard process and tailoring guidelines may be imposed. For example, the standard process may be over-specified and thereby over-constrain projects to the detriment of project success. The tailoring guidelines may not allow projects the flexibility they need to tailor standard processes to address their project-specific needs and priorities [Charette 2004]. In particular, these guidelines may not allow the process dexterity28 required to effectively address difficult project challenges and risks. To find ways to more effectively address risk (and opportunity) was one of the main motivations for the development of iterative and spiral approaches to product develop- ment—and more recently—of Agile methods. In today’s increasingly dynamic world, CMMI- based organizational process improvement approaches cannot rely exclusively on traditional project management approaches, waterfall-based project lifecycles, and heavyweight analysis me- thods. Conversely, standard processes may be under-specified and omit proven organizational and project practices. Similarly, the definition of standard processes may tend to treat all CMMI prac- tices equally (to earn a maturity level) and fail to recognize that some specific practices are critical to the business. Proper attention to these practices may add immediate value, but these practices can be lost in the noise of charging toward a maturity level rating. CMMI and Agile each brings something to the table on how to run the business that the other side should listen to and learn from. More dialogue would be healthy and beneficial for users, the pa- radigms, and the broader community.
28 We use the word “dexterity” in this sentence because we are trying to use a term other than "agility." 22 | CMU/SEI-2008-TN-003
6.3 CURRENT PUBLISHED RESEARCH There is a large and growing body of research on the adoption of CMMI [Elm 2007]. As was the case in the early days of CMMI (e.g., [Goldenson 2003]), the nature of the published research on the efficacy of Agile methods is largely anecdotal. Also similar to CMMI, there is no one way to use Agile methods. As already mentioned, CMMI addresses the what. Therefore, the studies of the costs and benefits experienced by organizations using CMMI actually report on how those organizations implemented a set of processes to meet the requirements and expectations set forth in selected CMMI process areas and not on the effectiveness of individual CMMI practices per se. Likewise, there are multiple ways to implement Agile methods. Because the methods offer some- what different how-to’s, some methods clearly work better in some situations than others. Well- intentioned implementations of Agile methods can fail to deliver a good product and uphold the people-centric values held dear by Agile and CMMI champions alike. In a case study written by Khaled El Emam, the employee morale at one organization actually became worse following the introduction of Agile methods until more explicit attention was given to addressing selected CMMI process areas [Chrissis 2007]. Many are too quick to pass judgment, declare victory, or declare failure. They expect instant re- sults and often assume that those results will persist ad infinitum. The first project using Agile or CMMI has to be a success or the newly tried approach is labeled a failure; equally, one success is assumed to represent an organization-level, institutionalized pattern. The reality is that neither is generally the case. There are few case studies in which the organization still exhibits the same success a few years later after the environment, projects, personnel, and management have changed. While formal studies of impact and benefits have been clearly solicited from and by the Agile community, there has been less energy behind systematic attempts to examine impacts (both suc- cesses and failures) that result from introducing Agile methods. It will take years to understand the efficacy of individual methods and the contexts in which there is high return on investment. The same understanding has taken years for CMMI and that journey continues. The problem with some of the existing research for both CMMI and Agile methods (or of new technologies in general) is that success stories are more likely to be reported at conferences than are failures and that research or reports based on such samples of convenience are therefore likely to be favorably biased. Furthermore, it is easier to claim success than to attribute success to specific, discrete practices separate from other environmental factors such as team composition, organizational culture, and customer interfaces29. Much like the adage, “It is impossible to prove a negative,” project failure can happen for many more reasons than those controllable by a project. Conversely, a project can claim that it did nothing more than change some practices or introduce some methods and thereby justify its claims that what it changed or introduced is the source of its success. In either case, the proof is not robust; therefore, most such research should be considered anecdotal.
29 Elm 2007 may be a notable exception. By design this recent research study targeted a broader range of pro- grams (including less successful programs) and tried to account for challenges in the program environment. An important finding of this study was that while higher process capability (or maturity) results in markedly better project performance, project challenge (i.e., technical, stakeholder and contract complexity; requirements, scope, and funding volatility) lessens the benefits of higher capability.
SOFTWARE ENGINEERING INSTITUTE | 23
6.4 THE HOPE IN RECENT TRENDS The public relations problem that CMMI (or any process or development model) faces is that each organization that uses CMMI affects many project participants (e.g., product developers) who are able to quickly spread the word of their painful experiences through existing developer communi- cation channels, irrespective of any benefit attained, whether immediate or eventual30. Nonetheless, recent experience reports indicate that CMMI and Agile methods can work well to- gether and result in business value [Sutherland 2007]. These reports provide one basis for the syn- ergies discussed in this report. Jeff Dalton (a CMMI appraiser and Agile leader [Dalton 2008]) remarked, “I’ve observed Agile organizations who fiercely apply little process nor produce valuable documentation in the name of being ”Agile.” Their rationale for the absence of process is found in the Agile Manifesto, which places a higher priority on working code than on project documentation. This prioritization, which is intended to shift the balance away from docu-centric traditional software development, is often interpreted as “no process or documentation” and can result in confusion and failure as systems become larger and more complex. When CMMI is applied appropriately and integrated success- fully, it can provide the infrastructure required to successfully scale the management and delivery of complex systems while maintaining the benefits of iterative and incremental development that Agile provides. Also, it’s stunning how few “Agile organizations” use Agile methods to design and deploy their set of processes” [Dalton 2008]. David Anderson (author of [Anderson 2004] and co-author of the APLN's Declaration of Interde- pendence) observes, “It’s a fair comment that many in the Agile community tend to lump tradi- tional and CMMI in the same category and use the terms synonymously. I was probably guilty of this back in 2002. However, I’ve come to the opinion that traditional really means waterfall or V- model software engineering with traditional project management (i.e., estimating, and tracking tasks in a dependency graph). In my view, CMMI transcends the definition or difference between Agile and traditional. But this view is not widely held or understood at this time.” An interesting recent development is the work of Jeff Sutherland (another Scrum founder) and others who at Agile 2007 presented how Scrum could build on an organization’s successful achievement of maturity level 5, and how CMMI and Scrum together may be the “magic potion” of software. High maturity organizations were able to manage change better and institutionalize the change. Hence, a change to Scrum worked better in a high maturity organization. The transi- tion was faster and institutionalized better. It would be fair, then, to assert that the perceived problems between Agile and CMMI stem not from Agile and CMMI being inherently inconsistent, but rather from a combination of mispercep- tions and the negative experiences with some organizations that imposed overly-heavy and re- strictive standard processes on all projects as part of using CMMI—approaches that may be effec- tive in larger, low-trust, and high-assurance contexts—but burdensome and paralyzing when used in smaller project situations. CMMI encourages organizations to use processes that best meet their business objectives. However, the business objectives of organizations vary widely.
30 Gary Wolf of Raytheon recounted, “When we first heard our organization had to get to maturity level 3, we were filled with negative feelings. Change is painful unless you are stuck in a fire. But if we had to improve our processes, we were resolved to do it the right way. After several months, somewhat to our surprise, life got bet- ter in the projects: there was less rework, fewer late nights, and more time for the right kind of work.” 24 | CMU/SEI-2008-TN-003
What is needed when using CMMI are standard processes (or sets of standard processes) that rec- ognize the broader spectrum of situations that may arise and that provide process solutions that collectively address all relevant project situations31. Were DeMarco and Lister in their 1999 book Peopleware: Productive Projects and Teams [De- Marco 1999] correct in asserting that high maturity organizations (in the era of CMM use) be- come conservative and risk averse? It requires a broadminded attitude in regards to the software engineering process to enable Agile method adoption. In general, Agile values align with a culture in which anyone can suggest a change (i.e., improvement) and there is less attachment to the sta- tus quo or established way of doing things. For example, some organizations place tremendous emphasis on level ratings and are loath to change anything perceived to challenge appraised processes upon which their ratings were based. Barry Boehm and Rich Turner describe the need for hybrid approaches that combine and balance strengths from both plan-driven and Agile approaches based on a risk assessment of a particular project’s characteristics (e.g., high risk of failure vs. an overwhelming need for speed to market) [Boehm 2002, Boehm 2003]. In his foreword to Boehm 2003, Alistair Cockburn (founder of the Crystal family of methodolo- gies [Cockburn 2004]) provides this summary of the book’s message, “If one has strong discipline without agility, the result is bureaucracy and stagnation. Agility without discipline is the unen- cumbered enthusiasm of a startup company before it has to turn a profit32.” We largely agree with these messages, but pragmatically it is difficult to know how much plan- ning and architecting (and related development activities) are needed to minimize project risk ex- posure; the answer would seem to be situation dependent. Add subsequent product evolution (and attendant product releases and associated stakeholder costs) and it would seem some significant discipline (in the sense of self-control) and experimentation is needed to properly assess which hybrid approaches work in which contexts and to generalize to software-related industries as a whole. We hope this report encourages both CMMI and Agile proponents (ideally, everyone in software- related industries) to do the following: 1. Recognize the value of each paradigm. 2. Resist common misperceptions. 3. Continue experimenting, learning, and reporting on what works and in which contexts.
31 There is also a need for more explicit coverage of Agile approaches within CMMI itself and in the guidance provided within CMMI training and for appraisal purposes. In particular, SCAMPI Lead Appraisers need to have a more versatile understanding and appreciation for Agile approaches, what they are, and how they can benefit the organization. This report can be considered a step in that direction. 32 Unfortunately, Balancing Agility and Discipline: A Guide for the Perplexed [Boehm 2003] is a misleading choice for a book title. The choice is not Agile or discipline because how can you continuously produce tested, working software in short time intervals without discipline? Perhaps a more accurate title might be Balancing Agile and Plan-Driven Approaches. CMMI can be used to address the full spectrum, with the risk assessment of a project’s characteristics described in [Boehm 2003] being the basis for creating the project’s defined process through appropriate process tailoring.
SOFTWARE ENGINEERING INSTITUTE | 25
The tendency to form camps, such as separate CMMI and Agile camps, is an inherent human cha- racteristic, even when it comes to software development. What is needed is a supertribe toward which all camps can feel an affinity [Immelman 2003]. This supertribe can facilitate continued experimentation, learning, and results reporting. Creating a supertribe that encompasses CMMI and Agile is the shared goal we have for this report—and we encourage you to join us in pursuing this goal. 26 | CMU/SEI-2008-TN-003
7 Problems Not Solved by CMMI nor Agile Part One of each CMMI model describes three aspects of development projects as (1) processes, (2) technology, and (3) people. CMMI makes no secret that it focuses on processes. Meanwhile, Agile methods clearly focus on people and allow people to determine technology and processes. Regardless of the focus of CMMI or Agile, neither can entirely prevent simple human error, de- parture of key personnel, impact of incompetent personnel, active or passive insubordination, or deliberate sabotage. Neither can prevent the effect of an employee’s personal life on his or her job performance. Each body of knowledge can have mechanisms to catch and manage such matters33, but neither can prevent them any more than either can prevent the local economy from creating a scarcity of qualified employees. Most organizations have experienced ill-fitting employees and the corresponding challenges in placing them or keeping them employed or happy. Most organizations have had to deal with em- ployee losses, loss of project funding, evaporation of technology resources or suppliers, supply and demand in the marketplace, premature obsolescence of core architectural components, and other issues beyond their immediate control. While derived and adapted abilities can positively influence the non-development aspects of product and service development, there are many aspects to development that neither CMMI nor Agile, as a first tier implementation, are meant to affect. The conclusion here is that neither CMMI nor Agile are panaceas to all that may hinder development. To address the issue of how to most effectively recognize and credit Agile approaches within a SCAMPI appraisal bears further research. Perhaps simply a special guidebook for appraising Agile organizational units or projects or a set of Agile alternative practices associated with CMMI goals accompanied by an overview of Agile values and practices would prove useful. Fundamentally, misuse is a problem for both CMMI and Agile. One of the original signatories of the Agile Manifesto recently wrote that he was hired to perform due diligence by a venture capital firm that had decided to invest only in companies practicing Agile methods. He found that of the companies he visited and examined, only three percent of those claiming to use Agile methods were actually doing so. (Whether another signatory would judge a similar percent is of course a different, though related issue.)
33 Such workforce-management problems may be better addressed by the People CMM [Curtis 2001], which con- tains practices for improving the selection, retention, training, motivation, alignment, and skills of an organiza- tion’s personnel.
SOFTWARE ENGINEERING INSTITUTE | 27
One of many forms of misuse includes aspects of practice introduction entirely unrelated to either Agile or CMMI. Introducing new practices poorly will likely result in failed implementations re- gardless of what is being introduced. During a panel session on Agile at SEPG 2006, an audience member asked the panel about an Agile implementation failure at his company. The context of the question was that the adoption of Agile was an executive edict with unrealistic expectations of return on investment, disconnected goals, absent the Agile principles, and devoid of the cultural necessities of long-term change success34. This situation was an organizational behavior matter that neither CMMI nor Agile could immediately resolve.
34 For articles about these kinds of situations, see www.Agilemanagement.net/Articles/Weblog/ThoughtsonEnterpriseAgile.html 28 | CMU/SEI-2008-TN-003
8 Conclusion Agile methods provide software development how-to’s, purposely absent from CMMI, which work well on small co-located projects. Meanwhile, CMMI provides the systems engineering practices often required on large, high-risk projects. CMMI also provides the process management and support practices (and principles) that help deploy and continuously improve the deployment of Agile methods in an organization regardless of organization or project size. A scaling limitation of several Agile methods has yet to be overcome with any consistency while still adhering to Agile principles. Adhering to Agile principles is inherently challenging given the context of large, long-term projects with geographically and organizationally dispersed project teams (a situation whose prevalence may be increasing). The mode of communication on large projects—in terms of spanning distance, time, and audiences—is necessarily a slower more in- volved endeavor than is possible with face-to-face, real-time, kinetic presence among project team members. Scrum provides an Agile “wrapper” for many project entities that allows for “Scrums of Scrums.” However, this approach still requires a strongly cohesive project team and can even wrap around non-Agile projects, thus begging the question as to the true agility of any project work being per- formed behind a “curtain” of an Agile-recognized approach.35 Although the complexities of large-scale projects require a more involved infrastructure, this fact is not a license to create unnecessary bureaucracies and unbalanced production at the expense of productivity. This creation of unnecessary bureaucracy occurs especially, but is not limited to, when CMMI is misused and a level rating is the principal expected outcome of a CMMI-based process improvement effort. As a result, over-engineering within process improvement activities is a common issue. Using process experts to create and deploy process improvement activities is not unusual, and often has the advantage of allowing these experts (and the project teams) to focus on their respec- tive tasks. But what makes this approach risky is that project personnel are frequently left out of process design activities36 and are disinclined or openly skeptical toward the adoption of process improvement activities. This situation is typical of some Six Sigma style approaches to process improvement as well. Quality process teams are created and given the responsibility for quality. As a result, developers abdicate their responsibility for improvement (or worse, for quality).
35 Some Agilistas question the inclusion of Scrum as an Agile development method as opposed to, perhaps, an Agile project management method. However, we have included it here because of its widespread recognition. 36 Whether due to the design of process improvement activities or due to project personnel being too busy to pro- vide meaningful inputs, each has the same negative effect.
SOFTWARE ENGINEERING INSTITUTE | 29
Using Agile principles when designing and selecting CMMI practices can create more acceptable and appropriate process definition activities. Furthermore, the incorporation of CMMI goals into the project activities of Agile teams can help make these teams more mature and/or capable at handling the continuity of projects. Implementing CMMI practices in an organization that uses Agile methods, while remaining faithful to Agile principles, should also yield well-conceived Agile practices that maintain a lean, high-trust set of activities sufficient for the development of the product. The following scenario suggests a hypothetical approach to using Agile ideas in CMMI-based process improvement. Someone must be identified as the product owner for any product produced by a development team. Thereafter, no artifact of process improvement should be generated without an explicit owner for that artifact. The appraiser is off-limits as a product owner (i.e. for every CMMI- inspired artifact, there must be a real owner on a project). Every task or practice on a project pro- duces an output. All process improvement-identified tasks and practices are an assumed part of the project and must be addressed as project activities. Either the output of a task serves process improvement purposes as is (i.e., an existing project activity already meets the need) or the output is specifically for process improvement—in which case someone must take ownership of the out- put. Depending on the project’s budget structure, someone may need to “pony-up” money to create the work-product. An output with no owner means there is no value in the output. Why should any project spend energy on outputs that have no value? CMMI does not require it either. Therefore, effort would need to be spent re-engineering the process until someone “cares” about the output. Until then, the output is waste. If those of us in both the Agile and CMMI camps would understand and accept our differences and explore the advantages of the other, we will find new ways of combining the ideas of both to bring improvement to a whole new level. Our challenge to CMMI and Agile proponents alike is to learn the value of the principles and technology of the other and begin to use them in their work. 30 | CMU/SEI-2008-TN-003
9 Epilogue: A Call to Action As authors and contributors to this report, we represent both sides of the perceived dichotomy and have deep experience in both camps, yet we universally agree that Agile methods and CMMI cannot only co-exist, but successfully integrate to bring substantial benefits to both Agile and tra- ditional software development organizations. In these closing paragraphs, we issue a “Call to Ac- tion” to experts of each approach. 9.1 A CALL TO ACTION FOR CMMI EXPERTS The SEI has always maintained that CMMI is methodology neutral and that, as a process model (and not a process in and of itself), CMMI can be applied to virtually any situation that involves the design, development, and delivery of systems and software. Information about CMMI has always been freely available to all those who have an interest, al- though the methodology, manifested in training classes, white papers, and technical reports, has not always been effectively presented with the appropriate messages to each constituent group. This presentation of information has sometimes created confusion and misperceptions within the different methodology camps. This confusion is caused, in part, by the biases of those who deliver the message, and those who receive it, as well as the methods used to present the information. In short, the message is just not being communicated or received clearly and all participants in the process share the burden (and consequences) of the inevitable result. The SEI cannot directly influence the biases of the receivers, but it can prepare the presenters and monitor the reactions to messages by various constituent communities. The presenters that can be influenced by the SEI include CMMI Instructors, SEI Partners, and SCAMPI Lead AppraisersSM. These presenters all can play a proactive role in the delivery of the message and the monitoring of reactions by the current and prospective CMMI user community. CMMI Instructors Members of the CMMI instructor community, particularly SEI staff and authorized CMMI in- structors, are often the “first line of defense” during an attendee’s sojourn into the world of CMMI. Authorized CMMI instructors are well trained in the CMMI model itself, but it is imposs- ible for them not to bring their own biases and experiences (for good and ill) into the selection and presentation of examples, ancillary discussions, and commentary. This additional material, at least based on feedback received from attendees, is often found to be as valuable or even more valuable as the authorized material provided by the SEI. CMMI instructors are encouraged37 to further engage the Agile community in their training by including examples from multiple types of organizations, including those that deploy systems us- ing Agile, waterfall, and other relevant methodologies, in (1) the eight existing Introduction to
37 A change request has been submitted asking the CMMI Product Team to include coverage of Agile principles and approaches, as appropriate, in CMMI models and in CMMI and SCAMPI training; and in its CMMI qualifica- tion, certification, and Partner program selection processes. (CRs are submitted as explained at http://www.sei.cmu.edu/cmmi/models/change-requests.html.)
SOFTWARE ENGINEERING INSTITUTE | 31
CMMI course exercises and (2) the course content itself. (This material is particularly relevant in the course modules related to Project Planning, Project Monitoring and Control, Measurement and Analysis, Requirements Development, Technical Solution, Verification, and Validation.) Instruc- tors are further encouraged to emphasize the distinction between an organization's processes in- tended for use in product development (or project management) from model process area practic- es intended as improvement practices to achieve improvement goals. CMMI Partners Firms that are SEI Partners are often the “face” of CMMI for many organizations, particularly those smaller, Agile companies who are considering the adoption of CMMI. SEI Partners often are capable of delivering multiple services, including those related to CMMI, and may already be working in some capacity with an organization when the desire to adopt CMMI first appears. In fact, they often play a key role in the discovery of that desire through the types of services they are marketing to their clients. They are thus in a unique position to nurture prospective CMMI user in their understanding of how the model can be implemented in their specific business envi- ronment. CMMI Lead Appraisers SCAMPI Lead Appraisers and Team Leaders have the greatest impact on acceptance of the mes- sage, and probably have a greater degree of responsibility in causing, though often inadvertently, some of the confusion in the software industry between the Agile and CMMI camps. Because ap- praisals are often conducted after an organization has spent time and money building and deploy- ing their set of organizational processes, a lead appraiser who does not have a solid understanding of Agile methods could, by virtue of conducting an appraisal in a way not conducive or sensitive to use of Agile methods, cause a great deal of harm to an organization adopting an Agile ap- proach. This problem, though uncommon, could be one reason an Agile organization may avoid CMMI altogether. The perception is that to be successful with CMMI, an organization must meet the ap- praiser’s expectations, not the CMMI model’s specific and generic goals. While this is an incor- rect assumption, it nevertheless persists in the Agile community. SCAMPI Lead Appraisers and Team Leaders are encouraged to further engage the Agile commu- nity by being knowledgeable of Agile methods now in use,38 including examples of these methods (where appropriate) when training appraisal teams and when discussing CMMI with sponsors and others in the organization. What About the SEI and the CMMI Model? Finally, the SEI itself can play a key role in clarifying and eliminating the perceptions that led to the confusion that exists in various systems and software development communities. The first place this clarification should occur is in the CMMI models themselves. Since the incep- tion of CMMI, and in all subsequent releases, models have included robust sets of examples in the informative material. These examples include suggested work products, subpractices, elabora- tions, and amplifications that, justifiably, seek to enhance the users’ understanding of CMMI, and to assist them in the practical implementation of CMMI specific and generic practices.
38 Agile methods now in use include Scrum, XP, FDD, and the SEI’s TSP. 32 | CMU/SEI-2008-TN-003
However, much of the informative material is steeped in the language of traditional systems de- velopment. As new informative material is added to the model in subsequent releases of CMMI, the CMMI Product Team will improve the communication and guidance provided by ensuring that the new informative material includes work products, elaborations, subpractices, and amplifi- cations illustrative of and compatible with Agile values and principles. 9.2 A CALL TO ACTION FOR AGILE EXPERTS Agile experts should recognize that the SEI acknowledges that “level mania” and imposing processes is misuse. Agile experts are encouraged to learn more about CMMI, to seek out and engage in productive conversations with CMMI experts who understand CMMI as it was in- tended. In doing so, Agile experts will learn the following about CMMI: When introduced correctly, CMMI provides transparency, learning, and reuse of what works. CMMI focuses on “what” not “how.” This short statement means that a software devel- opment approach must be introduced for an implementation of CMMI to be effective. CMMI is method and tool agnostic. Also, the use of iteration is just as consistent with CMMI as is the use of a waterfall approach. CMMI embodies sound systems engineering and software engineering principles appli- cable in project contexts that Agile approaches were not intended to address. CMMI provides an infrastructure for organizational learning and systematic improve- ment. CMMI provides a path for the effective use of processes, measurement, training, and im- provement. CMMI uses variation and modeling to reduce waste. CMMI fosters a routine that helps to identify and address gaps and lapses in attention to processes. Agile experts should become more familiar with CMMI practices that complement Agile practic- es addressing the following: non-functional requirements product architecture measurement and analysis planning for non-product oriented tasks risk management organizational learning By becoming more familiar with these practices, you can bring about organizational maturity and a broader and more effective deployment of Agile practices. Finally, Agile experts should help organizations implementing Agile do so in a way that discou- rages individual heroics by encouraging management of processes, organizational maturity, and long-range value. When faced with a CMMI implementation, Agile experts are encouraged to work to ensure that the organization's product development and project management practices are
SOFTWARE ENGINEERING INSTITUTE | 33
well defined and have strong ties to business and customer value. In essence, Agile experts are called to participate in ensuring that CMMI implementations in their organizations involve the appropriate staff and are relevant to their context. 9.3 THE BOTTOM LINE The first real step in the process of reconciliation has been taken and manifests itself in this report. We, as this report’s authors and contributors, represent those professionals in the CMMI and Agile communities who are passionate about the subject of Agile methods and CMMI and are concerned with the perceptions that the software industry holds on both sides of the issue. By chartering this group to collaborate and write this report, a demonstrable, concrete step into the exploration and understanding of the issue has been taken, and this Call To Action outlines steps that can be taken by experts in both communities that can begin to repair the perceptions and re- concile our two communities. The next step is up to you. All practitioners (whether working with an Agile approach or CMMI) have a responsibility to learn about other technologies (and not just CMMI or Agile) and make the best use they can of the ideas they encounter for the betterment of their project, their organization, and the larger software engineering community.
34 | CMU/SEI-2008-TN-003
10 CMMI and Agile Paradigm Comparison The following table compares and contrasts the CMMI and Agile paradigms along multiple di- mensions. This table provides a nice summary of concepts in a way that makes it easier to under- stand the viewpoints of both CMMI and Agile. Table 1: CMMI and Agile Paradigm Comparison Dimension CMMI Paradigm Agile Paradigm Organizational focus of The focus is on the organization or enter- The focus is on the project and team. Agile attention prise. Most benefit occurs when CMMI is methods can isolate (i.e., insulate) the implemented at organizational level so that project/team from the organization and still all functions and capabilities contributing to be effective. the development of products and services are addressed by the process improve- ment effort. Management Management plays an important role in Management is a coaching function (as ensuring project success. There is much opposed to traditional command-and- attention to project management, including control) that helps to eliminate barriers to ensuring that plans affecting the project progress. This view of management may are integrated with the project plan, de- be expanding as Agile approaches are pendencies are managed, coordination extended to address larger project con- issues are resolved, there is a shared vi- texts. sion for the work, and risk management is performed. Trust Some CMMI practices assume the need Agile methods originated from the recogni- to compensate for a low-trust environment tion that teams work best when they are (a key concern of the Agilistas). A low- composed of task-mature individuals oper- trust environment is often characterized ating in high trust groups. An Agile envi- by (1) safety and mission critical objec- ronment fosters high trust. tives, (2) high risk of failure, and most of all (3) multiple stakeholders who cannot be totally transparent relative to their in- tensions and commitments. Planning CMMI promotes macro (project-level) In Agile methods, there are multiple levels planning with an emphasis on establishing of planning, including high-level product a suitable defined process enabling the planning (including release planning) and at project to achieve its objectives. the beginning of each iteration, more de- Traditional planning approaches assume a tailed planning around the features to be long time horizon. Detailed plans are not addressed in that iteration. There is a required by CMMI but many of the exam- strong emphasis on flexibility and replan- ples encourage such an interpretation. ning as conditions change. Nevertheless, many who use CMMI also Use of Gantt charts (that map tasks to ca- use “rolling plans” (detailed only to the next lendar time) and graphs of task networks is iteration or quarter). There is emphasis on discouraged because the requirements on replanning and conditional change. which the tasks are based change fre- quently.
SOFTWARE ENGINEERING INSTITUTE | 35
Market/User Assump- CMMI is broadly beneficial but particularly Agile methods have the most benefit in an tion so when the target market becomes more emergent and not well-understood tar- mature and process innovation becomes a get/market. more important differentiator to organiza- tion success. Design Presumptions CMMI presumes the product architecture is Projects are most successful when corpo- selected or created in the early stages of a rate standard architectures are adopted project and is revisited when it becomes with flexibility applied as the project clear the selected architecture is no longer progresses. Some Agile methods downplay valid or when using an iterative lifecycle. the importance of architectures on the gen- eral principle that such are often prema- turely specified. Learning Learning happens in many ways, including: Learning happens at project/iteration levels, (1) organizational training based on ana- typically bottom-up and just-in-time (i.e., lyses of process and skill needs and priori- comparable to Kaizen). ties and the design of appropriate training vehicles; (2) through the development activities of the project itself (e.g., the eval- uation of proposed product requirements and solutions for feasibility and cost, sche- dule, quality, and risk impact); (3) through the planning and use of the processes on projects, measurements, and lessons learned are gathered and shared across the organization; (4) as part of a causal analysis process; and (5) through an anal- ysis of how well the organization’s and project’s quality and process performance objectives are being met by the processes in use. Perspective CMMI takes, has, and assumes a longer- Agile takes, has, and assumes a short to term view. medium-term view. Appraisals The SCAMPI method compares the organ- The desire of Agile practitioners is that ization’s processes against the practices of appraisals are made only by looking at CMMI to evaluate whether the organization results (i.e., customer satisfaction and other has implemented processes that achieve project outcomes). CMMI goals. Human Development CMMI has a limited people focus at the Agile has a team and individual focus (i.e., project level. There is an expanded people people over process). focus at the organizational level. CMMI There has been a tendency among many in does not advocate heroics, but instead the Agile community to advocate “just hire creating an environment in which people good people” with an implicit undercurrent can excel without heroics (i.e., an effective of developing “hero developers” within a process appropriately leverages people work-life balanced culture. and technology). 36 | CMU/SEI-2008-TN-003
Life-Cycle Emphasis CMMI has a strong “review-as-you- Agile methods employ concurrent devel- develop” emphasis. There is also a ten- opment, test iterations, and informal peer dency to read in CMMI the approach “veri- reviews of work products as necessary. fy often and validate at the end,” which is There is a tendency to see in Agile a “vali- NOT the correct way to understand what date often and verify second approach” CMMI is saying (e.g., see the introductory which may be a more correct reading. notes to the Validation process area).39 “Fail early, fail fast, and learn” are central to CMMI is consistent with an environment in Agile methods. The trade-off is seen as which there is a high cost of failure (see favoring the development of a useable and Trust above). The objective is to proactive- testable product vs. developing and analyz- ly avoid the high costs associated with ing requirements and product components. product failure (or time-to-market failure). The conclusion is that we cannot rely on A low cost of delay or low cost of failure is testing alone. Therefore, CMMI encourag- assumed. There is an assumption of in- es documentation, analyses, and reviews cremental delivery being viable. before product components are integrated Burn out can arise from repeated time- into a functional product. CMMI also en- boxed iterations, so special provisions courages frequent (early and mid-course) should be made for this eventuality (e.g., by validations to ensure the right product is revising the time-box duration based on being built. The project determines the experience feedback and by providing appropriate application and timing of these some slack around iterations). review and testing steps. Predictability Predictability is achieved for critical sub- Using time-boxed scope-limited iterations, processes (those process steps that are evolving designs/solutions, driving out de- leading contributors to or indicators of fects as early as possible, and failing quick- quality and process performance) through ly leads toward a predictable development statistical management (which involves, velocity. among other things, use of control charts). There is also an expectation of predictabili- Predictability is achieved at the project, ty of iteration delivery and scope of deli- iteration, or level by monitoring the perfor- very. The anecdotal evidence suggests that mance of critical subprocesses and period- Agile teams often struggle with predictabili- ically using the organization’s established ty of iteration scope, often de-scoping itera- process performance models, appropriate- tions (or sprints) in the final few days. This ly calibrated, to evaluate whether the de-scoping is rooted in a lack of statistical project is on track to achieving its objec- convergence of the velocity data combined tives. Finally, at the organizational level, with the analysis technique underlying the analysis of subprocess performance scope breakdown. across projects provides an increased understanding of the capability of the or- ganization’s processes and thus of which areas to target for innovation. Cost of Failure Historically, CMMI was developed in a Agile methods have flourished in a domain domain of high cost of failure. If a plane of low cost of failure or linear incremental crashes, the cost of failure is extremely cost of failure. Examples within this domain high. Examples within this domain include include Internet commerce, social network- the development of aircraft, weaponry, ing, and games development. spacecraft and safety-critical medical de- vices.
39 Verification confirms that the product meets its specifications; validation ensures it meets the needs of the user in the end-use environment.
SOFTWARE ENGINEERING INSTITUTE | 37
References/Bibliography URLs are valid as of the publication date of this document.
[Anderson 2004] David Anderson. Agile Management for Software Engineering: Applying the Theory of Con- straints for Business Results. Upper Saddle River, NJ: Prentice Hall Professional Technical Reference (ISBN: 0131424602).
[Anderson 2005b] David Anderson, Sanjiv Augustine, Christopher Avery, Alistair Cockburn, Mike Cohn, Doug DeCarlo, Donna Fitzgerald, Jim Highsmith, Ole Jepsen, Lowell Lindstrom, Todd Little, Kent McDonald, Pollyanna Pixton, Preston Smith, & Robert Wysocki. The Declaration of Interde- pendence. 2005. http://www.pmdoi.org/
[Cao 2008] Cao, Lan and Ramesh, Balasubramaniam. "Agile Requirements Engineering Practices: An Empirical Study." IEEE Software, January/February 2008.
[Carleton 2006] Anita D. Carleton & Tim Chick. “Extending the Team Software Process for Systems Engi- neering.” TSP Symposium. Long Beach, CA, 2006. http://www.sei.cmu.edu/tsp/sym2006- presentations/extendtsp.pdf 38 | CMU/SEI-2008-TN-003
[Charette 2004] Dr. Robert Charette, Laura M. Dwinnell, & John McGarry. “Understanding the Roots of Process Performance Failure.” CrossTalk: The Journal of Defense Software Engineering, Au- gust 2004.
[Chrissis 2007] Mary Beth Chrissis, Mike Konrad, & Sandy Shrum. CMMI: Guidelines for Process Integra- tion and Product Improvement, 2nd Edition. Boston, MA: Addison-Wesley (ISBN: 0321279670)
[Cockburn 2004] Alistair Cockburn. Crystal Clear: A Human-Powered Methodology for Small Teams. Boston, MA: Addison-Wesley (ISBN: 0201699478).
[Cusumano 1995] Michael A. Cusumano & Richard W. Selby. Microsoft Secrets: How the World's Most Power- ful Software Company Creates Technology, Shapes Markets, and Manages People. New York, NY: Free Press (ISBN: 0028740483).
[Swoyer 2005] Stephen Swoyer. “Agile Programming and the CMMI: Irreconcilable Differences?” Applica- tion Development Trends, February 2005.
SOFTWARE ENGINEERING INSTITUTE | 41
REPORT DOCUMENTATION PAGE Form Approved OMB No. 0704-0188 Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, search- ing existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regard- ing this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188), Washington, DC 20503. 1. AGENCY USE ONLY 2. REPORT DATE 3. REPORT TYPE AND DATES (Leave Blank) November 2008 COVERED Final 4. TITLE AND SUBTITLE 5. FUNDING NUMBERS CMMI® or Agile: Why Not Embrace Both! FA8721-05-C-0003 6. AUTHOR(S) Hillel Glazer, Jef Dalton, David Anderson, Mike Konrad, & Sandy Shrum 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION Software Engineering Institute REPORT NUMBER Carnegie Mellon University Pittsburgh, PA 15213 CMU/SEI-2008-TN-003 9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSORING/MONITORING HQ ESC/XPK AGENCY REPORT NUMBER 5 Eglin Street Hanscom AFB, MA 01731-2116
11. SUPPLEMENTARY NOTES
12A DISTRIBUTION/AVAILABILITY STATEMENT 12B DISTRIBUTION CODE Unclassified/Unlimited, DTIC, NTIS
13. ABSTRACT (MAXIMUM 200 WORDS) Agile development methods and CMMI (Capability Maturity Model® Integration) best practices are often perceived to be at odds with each other. This report clarifies why the discord need not exist and proposes that CMMI and Agile champions work toward deriving ben- efit from using both and exploit synergies that have the potential to dramatically improve business performance.
14. SUBJECT TERMS 15. NUMBER OF PAGES Agile, Agile methods, CMMI 49 16. PRICE CODE
17. SECURITY CLASSIFICATION OF 18. SECURITY CLASSIFICATION 19. SECURITY 20. LIMITATION OF REPORT OF THIS PAGE CLASSIFICATION OF ABSTRACT Unclassified Unclassified ABSTRACT UL Unclassified Standard Form 298 (Rev. 2-89) Prescribed by ANSI Std. Z39-18 298-102 NSN 7540-01-280-5500