Blog
Insights on AI, software development, and building better digital products
How We Review Code When Agents Write Most of It
In June we merged 16 pull requests into our own website. In the first 18 days of October we merged 167. Nobody reads that many diffs line by line. Here is how we review code now, and where a person still decides.
Terrance MacGregor
Fidelity is up and obedience is down!
As a developer, you will be silently fired if you execute on user stories exactly as written. It won't be on purpose and it will happen slowly over time. This is called "ticket obedience" and it is no longer rewarded in the age of AI.
Stop Building on Shifting Sand: Lock Your Object Model Before AI Touches It
AI-assisted projects turn to slop when there's no foundation. Lock two files — a plain-English object definition and a derived data model — and AI has to build on rock instead of shifting sand.
We Gave Every AI Agent a Job Title: It Changed Everything
We broke AI agents into domain-specific subagents with explicit ownership of routes, types, and business rules. The codebase got better overnight.
Giving the Frontend Its Own Lane
The frontend is always last and always rushed. Here's how a dedicated frontend agent with a fixed execution order eliminated merge conflicts and built UI against stable contracts.
Terrance MacGregor
We Built an Agent That Runs Our Entire Sprint While We Sleep
Sprint execution is project management with a keyboard. Here's how a sprint-runner agent handles every ticket end-to-end — no human in the loop.
Terrance MacGregor