From Orchestra Conductor to One-Person Tech Stack: returning to building with VS Code, Copilot and Claude
How AI helped me build product again with the strategic perspective of someone who has led complex systems.
1. Returning to the technical trenches
After years in project management and team coordination, code becomes something you supervise but rarely touch. You still understand the logic, the risks and the dependencies, but syntax starts to create an artificial barrier.
AI has removed much of that friction. You no longer need to remember the exact shape of a useEffect or every detail of a next.config.js to validate an idea. You can return to hands-on engineering with your mind still focused on business and architecture.
For a PM profile, that changes the game: you move from defining what should be done to doing it yourself. You no longer need to wait for someone else's sprint to test a hypothesis in Next.js or build a first working version.
2. Copilot: the necessary catalyst
It is only fair to say that without Copilot, returning to implementation would have been much slower. It was the bridge that made opening VS Code again feel natural.
Copilot acted like an always-available junior developer, reminding me how to write a React component, structure a fetch to Supabase or complete a Tailwind CSS class without breaking flow.
Its greatest success was turning natural language into a new kind of syntax. For those of us coming from product leadership, it allowed us to write code again using the business logic already in our heads.
3. Process maturity: why Claude became my senior lead
As the Lean Web project scales, the needs change. Copilot's linear assistance helps, but it falls short when you need reasoning, context and architectural consistency.
- Understanding the why: a PM needs coherence. Claude understands why we use the App Router, how that affects performance on Vercel and which decisions should stay stable.
- Managing complexity: when next-intl, multilingual routes or complex TypeScript types enter the picture, Claude behaves like a senior architect: it does not just suggest code, it organises the strategy.
- Less noise, more solution: the frustration of loops is replaced by the ability to read the full context and reach a useful answer on the first or second attempt.
4. My high-performance workflow
As a senior profile, I look for efficiency. I do not want to get lost in the tool; I want the tool to amplify judgment and speed.
- Conception: I define the logic, data flow and constraints with a PM mindset.
- Structure with Claude: I present the challenge, for example a subscription system with Stripe and Supabase, and get a database and component structure.
- Execution in VS Code: I implement with Copilot supporting fast autocomplete for Tailwind CSS and local logic.
- Refactoring: I use Claude to clean up code, review types and keep clean code standards.
Conclusion
The evolution from Copilot to Claude is not just a change of tool; it reflects a professional who knows what she wants. After 30 years, AI has allowed me to close the circle: to become again the creator I was at the beginning, but with the strategic vision of someone who has led complex projects.