← All work
WorkshopAI maturitySpec-driven developmentClaude CodeElectron
A hands-on AI workshop, in the team’s own codebase
A full-day workshop for a team building desktop software that drives physical hardware. We started with an AI-maturity assessment and finished with working agent tooling in their own repository.
- Context
- External client — hardware & desktop software company
- My role
- Consultant & workshop facilitator
- When
- 2026
- from diagnosis to working tooling in their repo
- 1 day
- dimensions in the baseline assessment
- 8
- assessment scale, from L0 to L4
- 5 levels
The situation
The team already generated around 40% of its code with AI, and felt “AI-enabled”. The assessment showed otherwise: volume isn’t maturity. The real bottlenecks were the product spec and QA, not typing code.
Format
Before: a prep call and a baseline score across eight dimensions of AI-assisted engineering.
Morning: strategy.
- A five-level maturity model, from traditional to agentic engineering.
- Spec-driven development and context engineering.
- A four-tier testing model, including a device simulator so agents can test without the physical hardware attached.
- Guardrails, and the one question that matters: what share of engineering work can be delegated reliably?
Afternoon: hands-on, in the client’s own codebase.
- A repository “context pack” so every AI session starts with the right knowledge.
- A PRD-writing skill and a QA-planning subagent.
- The first automated end-to-end test for their Electron app.
- An extractor that turns hardware protocol documentation into agent-ready context.
Why this format works
The workshop focused on leaving the team with working tooling in their own repository, built on their own bottlenecks, so they could continue using it after the session.