To pursue an approach of using both a CMM and agile methods, several issues must be addressed. Some issues are easily resolved through brief analysis. Other issues require major effort to satisfy the model.
Process Area Coverage
The most direct way to compare agile methods to the CMM-SW or the CMMI is to consider coverage of the key process areas in the CMM-SW or process areas in the CMMI. Mark Paulk has already published an analysis of extreme programming against the CMM-SW.2 We'll focus on the CMMI here, but the general approach remains the same. For each of the process area goals, can the practices of the agile method satisfy the goals? If not, is this a gap that needs to be augmented, or are these reasonable areas for tailoring the process area?
Figure 1 illustrates a possible mapping for XP and Scrum to the CMMI, assuming that they're used on a project for which the methods are a good fit.3 4 Whether the goals of the process area are satisfactorily addressed in any particular situation depends on the specific context. For example, XP's story approach or Scrum's tasks may be very effective at managing the requirements of an information system, but are probably not sufficient in a hard, real-time domain.
Figure 1 Mapping the CMMI process areas: ++ largely satisfies, + somewhat satisfies, - doesn't satisfy.
There are several interesting process categories to mention. It's probably not surprising that these agile methods don't say much about process management category in Figure 1. However, an organization's existing CMM approaches for identifying, piloting, and deploying processes can be used for introducing these techniques. What's interesting to note is that there appears to be a conscious effort on the part of the agile community to ensure that the processes and tools needed to successfully deploy agile methods are made available across the entire industry, not just to a particular organization. The web and other publishing mechanisms have been powerfully leveraged, so that anyone looking for the assets needed to put these approaches into practice can find them readily.5 6 To a large degree, the ubiquitous availability of these assets will help satisfy the process management category of the CMMI.
The real strength of the agile methods is in project management. Both XP and Scrum spend a great deal of effort in planning, although there's not much focus on long-range planning. In both techniques, long-range plans are fairly coarse-grained. The focus is instead on short-range planning. Watts Humphrey observed that the agile approaches reflect the axiom, "If you don't plan well, plan often."7
In Scrum, for instance, a product backlog is maintained of all pending tasks. The backlog is maintained in priority order. A 30-day sprint is organized such that the top items in the backlog are implemented. Coordination within the sprint is provided by a daily 15-minute Scrum meeting in which each team member explains what was done since the last Scrum meeting, what's planned before the next Scrum meeting, and what current obstacles exist. Progress against the backlog of tasks for the sprint is maintained so that overestimates and underestimates can be identified and corrective action taken. The style is very different than the work breakdown structure practices described in the CMMI, but still satisfies many of the goals.4
Other gaps can be satisfied with complementary practices. In some of these cases, the complementary practices must be tailored to fit with the agile method. For example, an independent group validating XP work products would need practices tailored to fit the XP team's rapid pace of iterationsa member of the validation group could be included on the XP team to ensure that the stories were testable, and to validate the tests. In other cases, there's no direct interaction between the complementary practices and the agile method, and less practice tailoring may be needed. For example, because agile development typically omits practices for process management, existing practices for introducing new processes should be suitable without much alteration.
Some of the gaps in Figure 1 reflect assumptions on the part of the method; for example, Scrum doesn't specify any particular configuration-management approach, because that's assumed to be part of the engineering practices. In some cases, an organization's existing processes provide the complementary process without any tailoring. In others, agile techniques may be combined.
One challenge for all process-management groups is finding the right level of detail when capturing processes. This is especially true for agile methods. By design, agile methods are simple to describe and understand, even through the work required to execute them may be very complex. Consider XP's pair programming and peer reviews. Traditional peer review processes define a formal approach for reviewing products, capturing defects, and collecting and analyzing related metrics. Peer reviews are satisfied differently when agile methods are used. The idea is that there should be continuous peer review, so all developers work in pairs at a terminal. This concept can be reasonably described in an organization's template for process descriptions. However, if one were to attempt to press a traditional peer review description onto pair programming and try to capture a detailed log of the defects captured during pair programming, the practice would get bogged down and lose its value.
Despite what many people think, it's not true that agile methods are without artifacts, although they're certainly less documentation-focused than traditional techniques. Still, this is an issue for organizations for whom the CMMI is the basis for rating their organization.
XP uses working code that customers can use and has tests that suites can validate automatically. These are easily verifiable products. Other artifacts are less formal and may prove more difficult to verify. Index cards are used frequently, as are whiteboards. However, some care and creative thought can retain these artifacts with minimal overhead. Some groups, for instance, capture their whiteboard sessions with a digital camera for later reference.
In other cases, such as in pair programming, no obvious artifact is generated by the practice. Assessment methods such as SCAMPI are much more documentation-oriented than earlier assessment methods. Artifacts are required in all areas, although interviews may provide additional evidence for affirmation8. In these cases it may be necessary to modify the process to make it auditable. In keeping with the agile approach, a simple approach should be used, such as a log of programming pairs.
There is a substantial burden of proof on an organization that wants to use agile methods and still satisfy the CMMI. The model even reads as follows (emphasis added):
When you use a CMMI model as a guide, you plan and implement processes that conform to the required and expected components of process areas. Conformance with a process area means that in the planned and implemented processes there is an associated process (or processes) that addresses either the specific and generic practices of the process area or alternatives that clearly and unequivocally accomplish a result that meets the goal associated with that specific or generic practice.9
A variety of CMM practices would need major tailoring or use of alternative practices. In some cases, it may even make sense to tailor the process areas. For example, the Technical Solution process area includes the goal, "Product components, and associated support documentation, are implemented from their designs." XP's approach doesn't design up front. Instead, simple software is built, complete with a complete test framework, and then refactored. The design emerges over the life of the project. This is certainly one of the more controversial aspects of XP, but if you're going to make it fit with the CMMI, you need to tailor the goal.
High-maturity organizations are supposed to be agile and able to incorporate new practices. The description of a level 5 organization says, "Optimizing processes that are agile and innovative depends on the participation of an empowered workforce aligned with the business values and objectives of the organization."9 Although other approaches may optimize for performance or reduced defects, agile methods are an approach for optimizing requirements flexibility and discovery.
However, less mature organizations may have more difficulty trying to blend a CMM and agile methods. Since the agile methods have no explicit role for process or quality assurance organizations, these groups may feel threatened. The agile evangelists may perceive the process groups as a barrier. Unfortunately, each party may be making erroneous assumptions about the other. While engineers on their own may be able to introduce agile methods on their projects, process groups have the skills and expertise for introducing practices more broadly in an organization. Some organizations have been persuaded to adopt agile methods after experiencing the pain of significant failures with other techniques. Process groups can play an important role in promoting agile methods without hitting rock bottom. Process groups have experience with conducting pilots and collecting data and can guide the agile advocates. Both the advocates for agile methods and process groups want to improve their organizations. Encouraged, empowered grassroots support can be a powerful asset to a process group.