Fixes · Field notes
Project rescues, written down.
Anonymised case studies of stuck AI projects, broken launches and codebases we got back on track.
Every engagement leaves behind a story worth writing down — the failing build that turned out to be a forgotten environment variable, the AI prompt that needed a smaller schema, the stuck launch that needed a 30-minute scope conversation more than another sprint of code.
This page collects those stories as they’re written up — anonymised where needed, technical where it helps. Use them as a sample of the kind of work and the kind of thinking we bring.
What we tend to fix
Patterns we see again and again.
→
“Our AI demo doesn’t work in prod”
Hard-coded prompts, no schema validation, no eval harness, no observability. The fix is rarely “a better model” — it’s the missing engineering around it.
→
“Our agency disappeared at 60%”
A handover audit, a working dev environment, the missing 40% built in 2–6 weeks alongside your team. We’ve inherited Laravel, Django, Rails, Node, .NET — most of them work fine once they’re given attention.
→
“AI is supposed to make us faster”
A team has Cursor, Claude Code, Copilot — and is shipping at the same pace as before. The fix is workflow, review discipline, and choosing what NOT to delegate to the agent. We train it.
Coming soon
Case study write-ups in progress.
First field notes go live in the coming weeks — agentic-AI rescue stories, AI-prompt postmortems, before-and-after engineering breakdowns. Until then, the pattern cards above are the honest summary of what we keep seeing.