Back to work
Own system
Product OS
A product operating system: structured specifications, AI agents and context engineering to take an idea to working software. The app is the implementation of that system.
What needed solving
Building software with AI agents works well in demos and poorly in real projects: context gets lost, decisions go unrecorded and every session starts almost from scratch.
Who it's for
Myself building my own products, and in the medium term any maker working with AI agents.
What I did
- Defining the specification model: how to describe a product so an agent can build it.
- Designing the orchestration flows: what each agent does and in what order.
- Building the application that implements the system.
- Testing it by using it to develop other real projects.
Decisions that mattered
- The specification is the source of truth: code can be regenerated, intent cannot.
- Explicit, versioned context instead of long conversations that degrade.
- The app and the method evolve together: every real project feeds the system.
What came of it
- An AI-assisted development flow I use daily on my projects.
- Several products built with the system, including this website.
Technologies
- AI agents
- Context engineering
- TypeScript
- Astro
- Structured specifications