The Bottleneck Moved: Why a Team Topology Cannot Absorb AI

The shift in constraint: AI's impact on SAFe's team topology

SAFe's cross-functional team model, which has been the foundation of value delivery for over four decades, is facing a significant challenge with the introduction of artificial intelligence (AI). AI has removed a forty-year-old constraint that made this team-based approach essential. As a result, the organization chart now appears to be holding the bottleneck.

The key axiom of SAFe is that cross-functional teams, comprising five to eleven people, are the smallest unit of value delivery. The framework provides a twelve-week planning cycle to manage dependencies and ensures that teams have the necessary scaffolding to work effectively. This model has been effective in distributing work across multiple individuals, requiring an interface layer to coordinate partially-informed team members.

However, with the advent of AI, the cost structure of software delivery has changed. Synthesis, writing code, writing tests, and designing alternative implementations of the same interface have all become significantly cheaper due to AI. Verification, the process of determining whether a candidate is correct, safe, and within organizational boundaries, has only seen a slight decrease in cost. This change has created a mismatch between the framework's planning system and the new reality.

The most expensive term in a delivery system is now verification and judgment, which AI can perform at zero marginal cost. Traditional team-based planning, which is optimized for synthesis costs, is no longer effective when verification becomes nearly free. The capacity to produce code and unit tests has been eliminated, but the planning system still relies on person-days to measure capacity and forecast PI planning.

This shift in constraint has led to two structural failures in the SAFe framework. First, the capacity to generate change has increased, while downstream integration and review processes remain unchanged. This results in a rise in local output and global work-in-process, leading to the illusion of progress. The organization experiences faster generation of change but slower integration and review, ultimately causing an increase in work-in-process and cycle time at the review gate.

Second, the forecasting layer of the framework is being corrupted by the increased generation of change. Story points and lines of code are inflating quickly due to AI assistance, creating a false sense of progress. However, downstream integration, architecture review, and security reviews still rely on human calendars with limited throughput. As a result, work-in-process continues to grow, and the organization reports faster shipping but slower review throughput.

In conclusion, AI has fundamentally changed the cost structure of software delivery, shifting the bottleneck from synthesis to verification and judgment. The SAFe framework, built on a team-based approach, is now ill-equipped to handle this new reality. The organization must adapt its planning and forecasting systems to accommodate the changing dynamics of AI-assisted software development.