AI-Assisted Engineering
AI-assisted engineering makes code faster to produce, but it does not remove responsibility for what reaches production. These field notes focus on the work that becomes more important when implementation gets cheaper: reviewing generated changes, evaluating probabilistic behavior, measuring outcomes rather than activity, and maintaining enough system knowledge to own the result.
The recurring question is not whether engineers should use AI. It is how an organization can gain speed without moving the bottleneck—and the risk—into review, integration, or production.
AI-Assisted Engineering Field Notes#
How AI Is Changing Engineering Team Structure
Teams are getting smaller and more senior while the pipeline that produces seniors is being cut. What is actually established, what is forecast, what would prove me wrong - plus an AI-era hiring rubric and the real cost of a junior residency.
Engineering Metrics in the Age of AI
Your board will ask whether AI made engineering more productive. Most available numbers answer a different question. A taxonomy separating diagnostic signals from performance measures, a paired scorecard, and the sentence to actually say in the meeting.
Evals Are the New Tests
In April 2025 OpenAI shipped a model update whose offline evals and A/B tests all looked good. It was sycophantic enough to roll back within days. A practical guide to building the measurement layer that would have caught it - with a working eval spec and grading rubric.
Code Was Never the Bottleneck
Your engineers are producing more code and your delivery cadence has barely moved. The waiting did not disappear - it moved downstream to review and absorption, and that is where the constraint always was. Includes a review-capacity audit.