Agility@Scale: Strategies for Scaling Agile Software Development

A common misunderstanding about agile software development approaches are that they're only applicable to small, co-located teams. Yes, it's much easier to be successful with small teams, and with co-located teams, and as a result agilists being smart people prefer to work this way. After all, why take on extra risk when you don't need to do so? But, sometimes reality gets in the way and you find yourself in a situation where you need a large team, or you need to distribute your team (see previous blog postings for strategies for distributed agile development), and you would still like to be as agile as possible. The good news is that it's still possible to be agile with a large team, although you will need to go beyond some of the popular "agile in the small" strategies to succeed.

Here are some disciplined agile strategies to succeed at large-team agile:

1. Question the need for a large team. Many times an organization will believe that they need a large team because their process is overly complex, because they're still organized for waterfall development, or simply because that's what they're used to. I've seen teams of 80 people doing the work of 20 as the result of over-specialization of job roles and all the bureaucracy required to organize and validate their work.
   
   
2. Do some initial envisioning. In order to succeed the team must work together towards the same goals. This is true for small teams but doubly true for larger ones -- without a common vision chaos will quickly ensue. You must gain this common vision on two fronts: you need a common business vision and a common technical vision. To gain the common business vision you must do some initial, high-level requirements envisioning and to gain the common technical vision some common architecture envisioning. This isn't to say that you need to take on the risk of detailed, up-front specifications but you must at least have a high-level understanding of the scope and technical solution in order to move forward effectively. So, expect to spend the first few weeks of your project doing this initial modeling.
   
   
3. Divide and conquer. You never have a team of 200 people, instead you have a collection of subteams that add up to 200 people. This is called having a team of teams.
   
   
4. Align team structure with architecture. The most effective way to organize the subteams is to have each one implement one or more components, and thereby to build your overall system as a "system of systems". This reduces the coordination required because the majority of the communication will be within the subteams themselves. You'll still need to coordinate the subteams, that will never go away, but you can reduce the overhead (and the risk) by being smart about the way that you organize the people. A common mistake is to organize around job function (e.g. having architects in Toronto, developers in Raleigh, testers in Bangalore, and so on). This increases communication overhead and risk because these people need to work together closely to get something built.
   
   
5. Project management coordination. Each subteam will have a team lead/coach, and these people will need to coordinate their work. There is often an overall project manager who leads this group. To coordinate the work within their subteam the team lead/coach will often have a daily meeting, in the Scrum method this is called a scrum meeting, where people share their current status and identify any problems they may be running into. To scale this effectively the team lead/coach attends a daily team coordination meeting, a scrum of scrums, where the same sort of information is shared at the overall team level.
   
   
6. Product owner coordination. Similarly, each subteam has a product ownder, also referred to as an "on-site customer", who is responsible for making decisions about the requirements and for providing information to the team in a timely manner. Sometimes a single product owner will work with several subteams. The product owners will get together at the beginning of the project to do some requirements envisioning to identify the initial scope and to start portioning the requirements between the subteams. Because the requirements between the subsystems are interrelated and should be reasonably consistent, the product owners will need to meet on a regular basis to share information, to negotiate priorities, and to resolve requirements-related disputes.
   
   
7. Architecture coordination. Each subteam will have an architecture owner, often a senior technical person and sometimes also in the role of the team lead/coach. These architecture owners will get together at the beginning of the project to do some initial architecture envisioning, based on the requirements envisioning efforts of the product owners. They will identify the major subsystems, and their interfaces, enabling the effective organization of the team into smaller subteams corresponding to the architecture. They will also get together regularly to evolve the architecture and to resolve any major technical issues.
   
   
8. System integration team. For complex systems, which is often what large teams work on, an effective system integration effort is critical to your success. Although this may be easy at first, as the overall system evolves the need for a subteam focused solely on this quickly becomes apparent. This not only supports the development efforts of the subteams, it also supports independent investigative testing.
   
   
9. Independent testing team. An independent testing team is common on mid-to-large size agile projects to enhance the testing efforts of the development subteams. This testing team will work in parallel to the developers, they get a new build on a regular basis (minimally each iteration, although more often is desirable), which they test in more advanced ways than what is typical with Test-Driven Development (TDD). For example, they often validate non-functional, quality of service (QoS) type requirements as well as technical constraints, things that often aren't captured well via user stories. They'll also do investigative testing to try to break the system by using it in ways not thought of by the product owners.
   
   
10. Some specialties reappear. On larger teams it can make sense to have some people be a bit more specialized than what we normally see on small agile teams. For example, it's common to see people in the role of agile DBA, tech writer, build master, or user experience (UE) professional. More complex systems often require people in these roles, although it still behooves these poeple to not be pure specialists but instead to be generalizing specialists with a wider range of skills. Also, recognize that the reintroduction of specialists can be a slippery slope back to the bureaucracy of traditional software development.

Further reading:

• Agile and Large Teams (DDJ Column, July 2008)
• System Integration on Large Agile Teams (DDJ Agile Newsletter, July 2008)
• Initial requirements envisioning
• Initial architecture envisioning
• Agile Testing Strategies (DDJ Column January 2007)

[Read More]

I'm happy to announce that a revised version of the Lean Development Governance white paper which I co-wrote with Per Kroll is now available.  This version of the paper reflects our learnings over the past few years helping organizations to improve their governance strategies.

There's a more detailed description of the paper here.

Contrary to popular belief, agile development teams do in fact model and yes, they even do some up front requirements and architecture modeling. Two of the best practices of Agile Modeling are Requirements Envisioning and Architecture Envisioning where you spend a bit of time at the beginning of the project doing enough initial modeling to get you going in the right direction. The strategy is to take advantage of modeling, which is to communicate and think things through without taking on the risks associated with detailed specifications written early in the lifecycle. In this blog posting I will focus on requirements envisioning, in a future posting I'll cover architecture envisioning.

The goal of initial requirements envisioning is to identify the scope of your effort. You need to do just enough modeling early in the project to come to stakeholder concurrence and answer questions such as what you're going to build, roughly how long it's going to take (give a range), and roughly how much it's likely to cost (once again, give a range). If you can get the right people together in the room, which can sometimes be a logistics challenge but not one that you couldn't choose to overcome, there are very few systems (I suspect less than 5%) that you couldn't initially scope out in a few days or a week. I also suspect that most of the remaining systems could be scoped out with less than 2 weeks of modeling, and if not then I'd take that as an indication that you're taking on too large of a project. I'm not saying that you'll be able to create big detailed specifications during this period, and quite frankly given the problems associated with "Big Requirements Up Front (BRUF)" you really don't want to, but I am saying that you could gain a pretty good understanding of what you need to do. The details, which you'll eventually need, can be elicited throughout the lifecycle when you actually need the information. A common saying in the agile community is that requirements analysis is so important for us that we do it every single day, not just during an initial phase. I'll discuss just in time (JIT) approaches to requirements modeling in a future posting.

To envision the requirements for a business application, you might want to consider creating the following models:

• High-level use cases (or user stories). The most detail that I would capture right now would be point form notes for some of the more complex use cases, but the majority just might have a name. The details are best captured on a just-in-time (JIT) basis during construction.
• User interface flow diagram. This provides an overview of screens and reports and how they're inter-related. You just need the major screens and reports for now.
• User interface sketches. You'll likely want to sketch out a few of the critical screens and reports to give your stakeholders a good gut feeling that you understand what they need. Sketches, not detailed screen specifications, are what's needed at this point in time.
• Domain model. A high-level domain model, perhaps using UML or a data modeling notation, which shows major business entities and the relationships between them, can also be incredibly valuable. Listing responsibilities, both data attributes and behaviors, can be left until later iterations.
• Process diagrams. A high-level process diagram, plus a few diagrams overviewing some of the critical processes, are likely needed to understand the business flow.
• Use-case diagram. Instead of a high-level process diagram you might want to do a high-level use case diagram instead. This is a matter of preference, I likely wouldn't do both.
• Glossary definitions. You might want to start identify key business terms now, although I wouldn't put much effort into settling on exact definitions. I've seen too many teams run aground on "analysis paralysis" because they try to define exact terminology before moving forward. Don't fall into this trap.

For small teams simple tools such as whiteboards and paper are usually sufficient for requirements envisioning. But what happens at scale? What if you're working on a large agile team, say of 50 people, 200 people (IBM has delivered software into the marketplace with agile teams of this size), or even 500 people (IBM currently has teams of this size applying agile techniques)? What if your team is distributed? Even if you have people working on different floors of the same building, let alone working from home or working in different cities or countries, then you're distributed (see my postings about distributed agile development). Suddenly whiteboards and paper-based tools (index cards, sticky notes, ...) aren't sufficient. You're still likely to use these sorts of tools in modeling sessions with stakeholders, but because of one or more scaling factors you need to capture your requirements models electronically.

In January Theresa Kratschmer and I gave a webcast entitled Agile Requirements: Collaborative, Contextual, and Correct which overviewed agile approaches to requirements elicitation and management, including requirements envisioning. We also showed how Rational Requirements Composer (RRC) can be used to electronically capture critical requirements information, enabling you to address the needs of large and/or distributed agile teams, while still remaining lightweight and flexible. I suspect that you'll find the webcast to be very illuminating and RRC something that you want to take a look at (the link leads to a trial version). Of course RRC can be used in other situations as well, but that's not what I'm focused on right now.

Teams which find themselves in regulatory environments will likely need to do more than just use RRC, as might very large teams. Regulatory compliance often requires more complex requirements documentation, which in turn requires more sophisticated tools such as DOORS or Requisite Pro, and I would consider using those tools in the types of situations that warrant it. One of the things that people often struggle to understand about agile approaches is that you need to tailor your strategy to reflect the situation at handle. One process size does not fit all, so you will end up using different tools and creating different artifacts to different extents in different situations. Repeatable results, not repeatable processes, is the rule of the day.


Further reading:

• Agile Requirements: Collaborative, Contextual, and Correct Webcast
• Rational Requirements Composer
• Rational's Agile Software Development Home Page (for webcasts, whitepapers, ...)
• Agile Requirements Modeling
• Requirements Envisioning: An Agile Best Practice
• Examining the Big Requirements Up Front (BRUF) Approach
• DDJ's 2008 Modeling and Documentation Survey, which found that agile teams are more likely to model than traditional teams

[Read More]

This article has been replaced by an official "Disciplined Agile Manifesto".

 

The text of the original article remains below.

---------------------------------------------------------------------------------------

I've recently been working with Mark Lines of UPMentors and we've had some interesting discussions around evolving the Agile Manifesto which I thought I would share here to obtain feedback.  Note that this is not any sort of official position of IBM, nothing in my blog is by the way (unless explicitly stated so), nor is it some sort of devious plot to take over the agile world (although if we did have some sort of devious plot, we'd make the exact same claim).  What we hope to accomplish is to put some ideas out there in the hopes of getting an interesting conversation going.

Over the past decade we’ve applied the ideas captured in the Agile Manifesto and have learned from our experiences doing so.  What we’ve learned has motivated us to suggest changes to the manifesto to reflect the enterprise situations which we have applied agile and lean strategies in.  We believe that the changes we’re suggesting are straightforward:

1. Where the original manifesto focused on software development, a term which too many people have understood to mean only software development, we suggest that it should focus on solution delivery.
2. Where the original focused on customers, a word that for too many people appears to imply only the end users, we suggest that it focus on the full range of stakeholders instead.
3. Where the original manifesto focused on development teams, we suggest that the overall IT ecosystem and its improvement be taken into consideration.
4. Where the original manifesto focused on the understanding of, and observations about, software development at the time there has been some very interesting work done within the lean community since then (and to be fair there was very interesting work done within that community long before the Agile Manifesto was written).  We believe that the Agile Manifesto can benefit from lean principles.

Our suggested rewording of the Agile Manifesto follows, with our suggested changes in italics.

Updating the Values of the Agile Manifesto

We are uncovering better ways of developing software by doing it and helping others do it.  Through this work we have come to value:

1. Individuals and interactions over processes and tools
2. Working solutions over comprehensive documentation
3. Stakeholder collaboration over contract negotiation
4. 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.

Updating the Principles behind the Agile Manifesto

1. Our highest priority is to satisfy the customer through early and continuous delivery of valuable solutions.
2. Welcome changing requirements, even late in the solution delivery lifecycle. Agile processes harness change for the stakeholder’s competitive advantage.
3. Deliver working solutions frequently, from a couple of weeks to a couple of months, with a preference to the shorter timescale.
4. Stakeholders and developers must work together daily throughout the project.
5. Build projects around motivated individuals.  Give them the environment and support they need, and trust them to get the job done.
6. The most efficient and effective method of conveying information to and within a delivery team is face-to-face conversation.
7. Quantified business value is the primary measure of progress.
8. Agile processes promote sustainable development.  The sponsors, developers, and users should be able to maintain a constant pace indefinitely.
9. Continuous attention to technical excellence and good design enhances agility.
10. Simplicity--the art of maximizing the amount of work not done--is essential.
11. The best architectures, requirements, and designs emerge from self-organizing teams.
12. At regular intervals, the team reflects on how to become more effective, then tunes and adjusts its behavior accordingly.
13. Leverage and evolve the assets within your organizational ecosystem, and collaborate with the people responsible for those assets to do so.
14. Visualize workflow to help achieve a smooth flow of delivery while keeping work in progress to a minimum.
15. The organizational ecosystem must evolve to reflect and enhance the efforts of agile teams, yet be sufficiently flexible to still support non-agile or hybrid teams. 

We’re agile – things evolve, including manifestos.  Looking forward to your feedback (add a comment).

 

Updates Since this Was First Published:

• February 2011: At the 10th Anniversary of the Agile Manifesto get together several people, including myself, brought up the idea of updating the Agile Manifesto.  Unfortunately the decision by the original writers of the manifesto was to adopt a very strict, non-agile change management approach where any changes to the manifesto would need the approval of all 17 of them.  So, pretty much zero chance of there ever being any changes (hence the lack of changes so far). 
• June 2012: This extension to the manifesto was retitled "A Disciplined Agile Manifesto" and published as in the book Disciplined Agile Delivery written by myself and Mark Lines.
• November 2012: Mark and I formed Scott Ambler+Associates.
• February 2013: The Disciplined Agile Consortium (DAC) is formed.  The mission of DAC is to evolve and support the Disciplined Agile Delivery (DAD) framework.
• April 2014: Official Disciplined Agile Manifesto posted on the Disciplined Agile Delivery (DAD) site.

Further reading:

• The Declaration of Interdependence (some other thoughts about, arguably, evolving the manifesto)
• Examining the Agile Manifesto
• Disciplined Agile Delivery website

 

 

 

The explicit phases of the Unified Process -- Inception, Elaboration, Construction, and Transition -- and their milestones are important strategies for scaling agile software development to meet the real-world needs of modern organizations. Yes, I realize that this is heresy for hard-core agilists who can expound upon the evils of serial development, yet these very same people also take a phased approach to development although are loathe to admit it. The issue is that the UP phases are like seasons of a project: although you'll do the same types of activities all throughout a project, the extent to which you do them and the way in which you do them change depending on your goals. For example, at the beginning of a development project if you want to be effective you need to do basic things like identify the scope of the project, identify a viable architecture strategy, start putting together your team, and obtain support for the project. Towards the end of a project your focus is on the activities surrounding the deployment of your system into production, including end-of-lifecycle testing efforts, training, cleaning up of documentation, piloting the system with a subset of users, and so on. In between you focus on building the system, including analysis, design, testing, and coding of it. Your project clearly progresses through different phases, or call them seasons if the term phase doesn't suit you, whether your team is agile or not.

The UP defines four phases, each of which address a different kind of risk:1. Inception. This phase focuses on addressing business risk by having you drive to scope concurrence amongst your stakeholders. Most projects have a wide range of stakeholdres, and if they don't agree to the scope of the project and recognize that others have conflicting or higher priority needs you project risks getting mired in political infighting. In the Eclipse Way this is called the "Warm Up" iteration and in other agile processes "Iteration 0".2. Elaboration. The goal of this phase is to address technical risk by proving the architecture through code. You do this by building and end-to-end skeleton of your system which implements the highest-risk requirements. Some people will say that this approach isn't agile, that your stakeholders should by the only ones to prioritize requirements. Yes, I agree with that, but I also recognize that there are a wide range of stakeholders, including operations people and enterprise architects who are interested in the technical viability of your approach. I've also noticed that the high-risk requirements are often the high-business-value ones anyway, so you usually need to do very little reorganization of your requirements stack.3. Construction. This phase focuses on implementation risk, addressing it through the creation of working software each iteration. This phase is where you put the flesh onto the skeleton.4. Transition. The goal of this phase is to address deployment risk. There is usually a lot more to deploying software than simply copying a few files onto a server, as I indicated above. Deployment is often a complex and difficult task, one which you often need good guidance to succeed at.

Each phase ends with a milestone review, which could be as simple as a short meeting, where you meet with prime stakeholders who will make a "go/no-go" decision regarding your project. They should consider whether the project still makes sense, perhaps the situation has changed, and that you're addressing the project risks appropriately. This is important for "agile in the small" but also for "agile in the large" because at scale your risks are often much greater. Your prime stakeholders should also verify that you have in fact met the criteria for exiting the phase. For example, if you don't have an end-to-end working skeleton of your system then you're not ready to enter the Construction phase. Holding these sorts of milestone reviews improves your IT governance efforts by giving senior management valuable visibility at the level that they actually need: when you have dozens or hundreds of projects underway, you can't attend all of the daily stand up meetings of each team, nor do you even want to read summary status reports.

These milestone reviews enable you to lower project risk. Last Autumn I ran a survey via Dr. Dobb's Journal (www.ddj.com) which explore how people actually define success for IT projects and how successful we really were. We found that when people define success in their own terms that Agile has a 71% success rate compared with 63% for traditional approaches. Although it's nice to that Agile appears to be lower risk than traditional approaches, a 71% success rate still implies a 29% failure rate. The point is that it behooves us to actively monitor development projects to determine if they're on track, and if not either help them to get back on track or cancel them as soon as we possibly can. Hence the importance of occasional milestone reviews where you make go/no-go decisions. If you're interested in the details behind the project, they can be found at http://www.ambysoft.com/surveys/success2007.html .

Done right, phases are critical to your project success, particularly at scale. Yes, the traditional community seems to have gone overboard with phase-based approaches, but that doesn't mean that we need to make the same mistakes. Let's keep the benefit without the cost of needless bureaucracy.[Read More]

My new paper Scaling Agile: An Executive Guide is now available.  As the title suggests the paper overviews how to scale agile strategies to meet your organization's unique needs. 

The executive summary:
Agile software development is a highly collaborative, quality-focused approach to software and systems delivery, which emphasizes potentially shippable working solutions produced at regular intervals for review and course correction. Built upon the shoulders of iterative development techniques, and standing in stark contrast to traditional serial or sequential software engineering methods, agile software delivery techniques hold such promise that IBM has begun to adopt agile processes throughout its Software Group, an organization with over 25,000 developers. But how can practices originally designed for small teams (10-12) be “scaled up” for significantly larger operations? The answer is what IBM calls “agility@scale.”

There are two primary aspects of scaling agile techniques that you need to consider. First is scaling agile techniques at the project level to address the unique challenges individual project teams face. This is the focus of the Agile Scaling Model (ASM). Second is scaling your agile strategy across your entire IT department, as appropriate. It is fairly straightforward to apply agile on a handful of projects, but it can be very difficult to evolve your organizational culture and structure to fully adopt the agile way of working.

The Agile Scaling Model (ASM) defines a roadmap for effective adoption and tailoring of agile strategies to meet the unique challenges faced by a software and systems delivery team. Teams must first adopt a disciplined delivery lifecycle that scales mainstream agile construction techniques to address the full delivery process, from project initiation to deployment into production. Then teams must determine which agile scaling factors – team size, geographical distribution, regulatory compliance, domain complexity, organizational distribution, technical complexity, organizational complexity, or enterprise discipline, if any — are applicable to a project team and then tailor their adopted strategies accordingly to address their specific range of complexities.

When scaling agile strategies across your entire IT organization you must effectively address five strategic categories — the Five Ps of IT: People, principles, practices, process, and products (i.e., technology and tooling). Depending on your organizational environment the level of focus on each area will vary. What we are finding within many organizations, including IBM, is that the primary gating factor for scaling agile across your entire organization is your organization’s ability to absorb change.

Recently I visited a customer who had adopted Scrum. They were a few sprints, what Scrum calls iterations, into the project and were running into some difficulties. Although I was primarily brought in to educate senior management on disciplined agile software development, I was also asked to sit in on the team’s daily stand-up meeting so that I could hopefully provide some suggestions as to how to address the problems they were running into.

Their work area was fairly typical. They had some whiteboards which they were using for project planning and tracking, with sticky notes to indicate what work had been taken on by each team member. The current status of the task (not yet started, in progress, and completed) was indicated by putting each sticky note in a corresponding column for the status and corresponding column for the team member. This allowed everyone on the team to easily share their status and to see the status of everyone else. On the sides were sketches of the architecture as well as some business oriented models. In addition to Scrum the team had adopted several practices from Agile Modeling, in this case they had done some initial requirements envisioning and architecture envisioning, as well as practices from Extreme Programming (XP) for construction. In short, they had followed a fairly common strategy of combining practices from various agile methods.

This would have worked perfectly fine if they had tailored the practices to reflect the situation that they were in, but instead they adopted them "straight out of the book". First, the team was distributed, with most of the team in the location that I was visiting but some people located in two other distant cities. Therein was the source of most of their problems. The people at the other two locations weren’t getting much value out of the daily stand-up meetings, even though they would dial in, because they couldn’t see the project status information. Although people at this location were trying their best to represent these distant people in the daily stand-ups it wasn’t working well – their status information wasn’t being kept up to date and for some people it was a bit of mystery as to what they were actually working on at all.

This team also had 30 people in it, which isn’t a big deal although it can stretch the limits of the simple modeling and planning tools (in this case paper and whiteboards) that they were using. Because the team was larger they were investing a fair bit of time creating burn down charts at both the iteration/sprint and project levels. One of the unfortunate implications of using manual tools for project management is that any associated metric/status reporting in turn becomes manual as well. Considering how the agile community is so concerned with working efficiently, I find it comical that we have a tendency to overlook our own potentially unnecessary bureaucracy such as this.

The problem was that the team was applying strategies, in this case using sticky notes and whiteboards to capture the detailed iteration plan, applying similar strategies to capture key models, and were verbally relaying of status information between sub-teams. There are perfectly fine strategies for smaller co-located teams, but not so good for large or distributed teams. The solution was to recognize that they were in an Agility@Scale situation and needed to tailor their approach to reflect this fact. In this case they needed to forgo some of the manual tools and instead use electronic tooling such as Rational Team Concert (RTC) to share information across disparate locations, in particular the work assignment and corresponding status information. RTC also creates common agile reports such as burn-down charts based on the activities of the developers, providing accurate (nearly) real-time information while removing the burden of status reporting. The RTC project dashboard does more than just this, to see an actual example of one visit www.jazz.net to see the dashboard for the RTC development team itself. You can also see their actual work item list too, a more advanced version of Scrum’s product and sprint backlogs.[Read More]

Test-driven development (TDD) is a common agile programming technique which has both specification and validation aspects. With TDD, you specify your software in detail on a just-in-time (JIT) basis via executable tests that are run in a regression manner to confirm that the system works to your current understanding of what your stakeholders require.

TDD is the combination of test-first development (TFD) and refactoring. With TFD, you write a single test (at either the requirements level with customer/acceptance tests or the design level with developer tests) and then you write just enough software to fulfill that test. Refactoring is a technique where you make a small change to your existing code to improve its design without changing its semantics.

TDD offers several benefits:1. It enables you to take small, safe steps during development, increasing programmer productivity.2. It increases quality. Agile developers are doing more testing, and doing it more often, than ever before. We're also fixing the problems that we find right on the spot.3. It helps to push validation activities early in the lifecycle, decreasing the average cost to fix defects (which rises exponentially the longer it takes you to detect them).4. Through single sourcing information, by treating tests as both specifications and as tests, we reduce the work required, increasing productivity.5. We leave behind valuable, up-to-date, detailed specifications for the people who come after us. Have you ever met a maintenance programmer who wouldn't want a full regression test suite for the code that they're working with?

But TDD isn't perfect. Although TDD is great at specifying code at a fine-grain level, tests simply don't scale to address higher level business process and architectural issues. Agile Model Driven Development (AMDD) enables you to scale TDD through initial envisioning of the requirements and architecture as well as just-in-time (JIT) modeling at the beginning and during construction iterations. To scale requirements-level TDD, you must recognize that customer tests are very good at specifying the details, but not so good at providing overall context. High-level business process models, conceptual domain models, and use cases are good at doing so, and these work products are often created as part of your initial requirements envisioning and iteration modeling activities. Similarly, to scale design-level TDD you must recognize that developer tests are very finely grained but once again do not provide overall context. High-level architecture sketches created during envisioning activities help set your initial technical direction. During each construction iteration, you'll do more detailed design modeling to think through critical issues before you implement them via TDD.

You also need to scale the validation aspects of TDD. TDD is in effect an approach to confirmatory testing where you validate the system to the level of your understanding of the requirements. The fundamental challenge with confirmatory testing, and hence TDD, is that it assumes that stakeholders actually know and can describe their requirements. Therefore you need to add investigative testing practices which explore issues that your stakeholders may not have thought of, such as usability issues, system integration issues, production performance issues, security issues, and a multitude of others.

For further reading, I suggest:1. My article "Introduction to TFD/TDD" at http://www.agiledata.org/essays/tdd.html which overviews TDD.2. My February 2008 column in Dr. Dobb's Journal entitled "Scaling TDD" at http://www.ddj.com/architect/205207998 which explores this issue in detail. 3. Andrew Glover's article "In pursuit of code quality: Adventures in behavior-driven development" at http://www.ibm.com/developerworks/java/library/j-cq09187/ which describes a new-and-improved take on TDD called BDD.[Read More]

• Examining the Agile Manifesto 
• Reworking the Agile Manifesto for Enterprise Agile
• Scaling Agile: Start with a Disciplined Foundation
• Scaling Agile: The Software Development Context Framework
• Disciplined Agile Delivery (DAD) home page
• Generalizing Specialists: Improving Your IT Skills