Skip to content

AI-Assisted Software EngineeringA Methodology That Actually Works

Start with a conversation. Document thoroughly. Execute systematically. Ship production-ready code.

The Process โ€‹

  1. Brainstorm with Opus โ€” Real conversation in a Claude Project. Discuss scope, tech stack, architecture. Multiple turns, not one prompt.
  2. Generate foundation docs โ€” README, Architecture, rules, sprint plan, task specs. Thorough enough that AI can execute autonomously.
  3. Execute in focused tasks โ€” One task per conversation. Plan mode first. Test after. Score confidence. Close. Next.
  4. Know when to start fresh โ€” Side tasks, circular debugging, long conversations โ€” write a task doc and start a new conversation.
  5. Audit between phases โ€” Fresh AI reviews your code. Fix what it finds before continuing.

Who This Is For โ€‹

  • Developers who want AI to accelerate their work without producing garbage
  • Founders validating ideas before burning runway
  • Team leads establishing standards for AI-assisted development

Real Example โ€‹

The VH Conference Toolkit โ€” a suite of open-source tools for event professionals โ€” was built using this methodology. Browse its repo to see thorough architecture docs, strict development rules, sprint-based task specs, and architectural decision records in action.

Get Started โ€‹


Need a hand getting started?

Whether you're trying AI-assisted development for the first time or tightening up a workflow you already have, I can help. I build apps, and I teach founders and teams to build their own โ€” from a one-off review of where you're stuck to hands-on help shipping the thing.

Written by Richard Osborne ยท Digital Bricks