Models, particularly those created early in the lifecycle, are little better than speculative promises regarding what you intend to build or how you intend to build it. Promises are nice, but validation that you're fulfilling those promises is even nicer. Although many agilists have adopted a test-first or test-driven approach to development, an incredibly good practice, as I described in "Scaling Test-Driven Development" (www.ddj.com/architect/205207998) this is at best a confirmatory approach to testing. The goal of confirmatory testing is to validate that you've built the system to your understanding of the requirements. Test-driven development (TDD) is effectively the agile equivalent of testing against the specification, but that assumes that the requirements are identified and understood. If you're not doing TDD you've got an even bigger challenge.
The point is that TDD is a great start but that you need more. As Figure 1 shows, disciplined agile teams include an independent testing effort that runs in parallel to development. The basic strategy is that the agile team should deploy a working build of their system to an independent test team on a regular basis, at least once an iteration, so that they can perform investigative testing and other higher forms of testing. This independent testing effort addresses concerns that are typically addressed by traditional teams during the testing phase at the end of the lifecycle. By doing this testing in parallel with development, and by reporting issues back to the development team quickly, you can reduce the average cost of addressing any defects and often shorten the overall time of your project. The development team treats the identified defects as potential requirements to be put onto their prioritized work item list and eventually addressed through TDD.
The independent test team is small, perhaps one independent tester for every 10 people on the development team, because the development team should be addressing confirmatory testing (which is the majority of the testing work). The independent testers are highly skilled testing specialists, perhaps the top 5 to 10 percent of the testing community, who have access to sophisticated testing tools. They will use these tools to try to break the system, and this is often around the types of issues addressed by NFRs and constraints.