The increase in developer productivity has not led to a comparable acceleration of releases. The reason is that the bottleneck has moved higher up the stack: into the area of requirements formalization and result verification.
With the advent of AI coding, teams expected a linear acceleration in delivery. In practice, only one stage sped up—the coding itself. After that, the system starts to degrade: the review queue grows, approval times increase, and correctness verification becomes more complex. Data from Faros AI shows a typical pattern: with a 21% increase in the number of tasks and almost double the number of merges, review time increased by 91%. This is a classic case of local optimization, which increases pressure on adjacent stages. Code is no longer the bottleneck; understanding—what exactly needs to be built and how to ensure it works correctly—has become the scarce resource.
In response, the work model itself is changing. The “white box” approach—reading every line—does not scale when thousands of lines are generated per hour. The “black box” approach—accepting results without verification—is not applicable in production. Practice is shifting toward the “grey box”: the engineer is responsible for two control points—formulating the specification and verifying the result through observable artifacts. This shifts the focus from implementation to intent. The trade-off is clear: we lose detailed control over the code, but gain speed, provided we can formalize requirements and verify system behavior through testable criteria.
Implementing this approach requires process restructuring. The specification becomes the central artifact: it must be precise enough for the agent to execute, and verifiable enough for the result to be validated without reading the implementation. In practice, this means:
– explicit acceptance criteria and coverage of corner cases
– documenting architectural decisions as part of the specification
– shifting reviews from “how it’s written” to “is it proven to work”
– using metrics (DORA) and prioritization (RICE) to eliminate bottlenecks in DevEx
This also affects team structure. If the main value lies in aligning intent, then communication is no longer overhead—it becomes the core work. Small teams benefit not because of lower coordination overhead, but because they reach shared understanding faster and formulate specifications more precisely.
As a result, the very “contract of responsibility” changes: the engineer is no longer the author of every line, but remains the owner of the outcome. Control is elevated one level higher—from code to intent. Metrics confirm the shift, but the final effects depend on process maturity: without investment in specification and verification, accelerated coding merely redistributes delays rather than eliminating them.