The old picture
A year earlier, AI mostly completed lines of code. It saved small amounts of time, often enough to matter in aggregate, but it did not change the shape of an engineer’s day.
The more important opportunity was sitting around the code: reading dependency changelogs, triaging errors, loading context from unfamiliar parts of the system, and deciding which work deserved attention first.
What changed
We moved in steps. Multi-file tools made broader context practical. Specialized sub-agents replaced one overloaded prompt. A plan-then-execute practice let engineers redirect the approach before code existed. AI became the first pass on pull-request review and dependency-risk analysis.
The clearest change arrived when an overnight workflow began reading the prior day’s Sentry alerts, checking Linear for existing work, creating missing tickets, ranking risk, and identifying the fixes that were safe enough for an automated pull request.
The morning started with judgment instead of context loading.
My contribution
I helped create and evolve the workflows, context, agent responsibilities, and guardrails that made this practical. The work was as much about defining boundaries as adding capability:
- Which tasks could proceed without a person.
- Where a plan needed review before execution.
- What blast radius an agent was permitted to operate within.
- Which evidence a human reviewer needed to make a fast, informed decision.
The outcome
AI became involved in planning, implementation, review, debugging, and operations. A two-to-three-person team operated at throughput historically associated with a larger team, based on internal analytics.
The model was not the durable advantage. The compounding work was the context, guardrails, and team habits built around it.
What remained difficult
Every new capability changed the boundary again. Patterns that felt advanced one quarter could look dated the next. The system needed continuous evaluation, and speed only remained useful while the team preserved trust and human judgment.
