A deep-tech engineering organisation that did not need another slide deck on prompting. They needed AI inside the workflows their developers already live in.
Starting point
Engineering leadership saw scattered, individual AI usage: a few power users, no shared standard, no reusable tooling. Classic training rounds had explained the tools but had never changed how a single pull request was reviewed.
How we ran it
- Challenge selection straight from the engineering backlog, not from a curriculum
- Mixed squads of senior and junior engineers, each with a build coach
- Build on real repositories and real internal data from hour one
- Closing pitch in front of engineering leadership with a go / no-go decision
Built during the week
- An AI code-review assistant plugged into the existing review flow
- Reusable prompt and agent patterns for recurring engineering tasks
- A documented internal playbook so new team members can pick it up
Outcome
- Broad engineering-wide adoption instead of a handful of power users
- Review turnaround shortened on the workflows covered by the assistant
- A working tool the team kept using after the week, not a certificate
Our engineers built a code-review assistant in two days that the whole team still uses. No training ever did that.
Engineering lead · NavVis