AI for engineering teams
Coding assistants with a working method behind them: specify, generate, review, evaluate.
Where the problem is
The whole team has Copilot. Speed goes up, review does not keep pace.
Where the problem is
The whole team has Copilot. Speed goes up, review does not keep pace.
- Generated code gets less review than hand-written code.
- Everyone uses the assistant differently.
- There is no shared view on what to delegate and what not to.
- Tests and documentation are still the bottleneck.
How I approach it
A way of working with assistants that keeps code quality intact.
- Training on GitHub Copilot and coding agents.
- Up-front specification as a team practice.
- Automated tests and review over generated code.
- Prompts and skills shared across the team.
- MCP integrations to connect internal systems.
- Observability of cost, latency and quality.
Three ways to see it working
Development flow
From issue to pull request with specification, tests and assisted review.
Documentation
Technical documentation generated from the repository and updated on each release.
MCP servers
Internal systems exposed to the team's assistants in a controlled way.
From the first meeting to real use
In phases, with a decision at the end of each one. There is no need to commit to a year up front.
Phase 1
Assessment of the current flow and its friction points.
Phase 2
Hands-on training on the team's own codebase.
Phase 3
Review, tests and automated evaluation.
Phase 4
Measuring quality, cycle time and team satisfaction.
Indicators we agree on before starting
No invented percentages: the metrics are set at the start and measured with your own data.
Cycle time
From issue to production.
Test coverage
Over generated code as well.
Shared judgement
The team uses the assistants in similar ways.
What it is built with
Other situations
Does this problem sound familiar?
Tell me how it shows up in your company and I will tell you whether it fits and where I would start.