This started with AsNoTracking() and ended with a race condition I caused as the reviewer.
Six months ago I rolled out GitHub Copilot to my team. I took notes the whole way — incidents, data, architecture decisions, governance failures. It turned into five articles.
This is the sixth. The retrospective.
Here's what I actually got wrong:
❌ The governance policy I spent two weeks writing — the team told me it was useless before I finished it
❌ I didn't measure reviewer load, only developer speed
❌ I waited five months to name something the team needed to hear at month two
❌ I was overconfident in month one — Copilot is very good at making you feel like everything is fine
And the finding I hadn't planned to track turned out to matter most:
Developers didn't say they felt more confident.
They said they felt less exhausted.
One developer said something in a retro I haven't stopped thinking about:
"I've started to feel anxious when Copilot doesn't have a suggestion."
That's the moment the whole series was really about.
The article covers:
→ What actually changed in four months of data
→ Why governance-as-document always fails
→ The three-layer system that replaced it
→ A decision tree for when to accept a suggestion
→ The skill atrophy problem nobody talks about
→ Everything I'd do differently from day one
Top comments (0)