Discipline at Scale
There are several "Agile at Scale" practices that require a level of discipline rarely seen on traditional teams. First, Agile teams not only do a bit of initial architecture modeling at the beginning of a project, they also do the work necessary to prove that the architecture works early in the lifecycle. This practice, common in the Unified Process, reduces technical risk because it doesn't assume that the architecture works simply because the architects say that it will. It requires greater discipline than the traditional approach of detailed up front modeling because it becomes very clear very quickly whether the team actually has what it takes to deliver a working system.
Second, identifying quality of service (QoS) requirements, such as security and usability issues, and then convincing business stakeholders to weave them into their prioritized work item stack requires significant skill and discipline. It would be undisciplined not to address these requirements at all, but almost as bad would be to invest months at the beginning of a project to address them before delivering anything of business value. Yes, it's a lot of fun to work on cool technical frameworks, but that isn't what our business stakeholders are actually paying us for.
The third scalability issue surrounds software process improvement (SPI). The Agile approach is to regularly, at least once an iteration, take the time to consider how effectively the team is working together and whether something can be done to improve things. This continuous SPI approach enables Agile teams to take advantage of their insights right away, improving their productivity throughout the project. It would be undisciplined not to attempt SPI at all or to wait until the end of a project to identify "lessons learned" in project postmortems that do little more than give people a sense of closure after their traditional death march.