Solving the New Bottleneck: Why AI Coding Tools Aren't Increasing Sprint Velocity

When the constraint moves, you have to move with it

Intuit Engineering Director Eric Anderson highlights a shift in software economics. He notes that the incremental cost of a line of code is now among the most inexpensive activities in development.

Why This Matters

Organizations are upgrading their technical ‘engine’ with AI tools but maintaining ‘road’ processes—such as rigid sprint structures and waterfall-style design handoffs—designed for an era where code generation was the primary constraint. This creates a systemic mismatch where increased individual output simply piles up in front of outdated review gates and cross-functional sign-offs, nullifying productivity gains.

Key Insights

• Theory of Constraints (2026): Improving one constraint without addressing the next merely creates inventory/work piling up before the new bottleneck.
• Requirement Discipline: When code is cheap, underspecified requirements lead to rapid but incorrect AI output, increasing rework costs.
• Review Surface Area: Increased output doubles the burden on senior engineers for architectural oversight and QA, shifting the bottleneck from writing to reviewing.

Practical Applications

• Use case: Intuit shifting toward PMs and engineers co-developing features in real time rather than using finished specs as prerequisites.
• Pitfall: Treating discovery and requirements as check-boxes on the way to coding, resulting in higher cycle times due to rework.

References:

• https://stackoverflow.blog/2026/06/18/the-new-bottleneck/