The context problem: 25 days stuck on MakeCode, fixed in hours
I spent 25 days failing to add sign-in and storage to Microsoft MakeCode with an AI assistant. Changing what I asked for fixed it in hours.
Abdullah Abdul Samad, . A 1 minute read. First posted on LinkedIn in February 2026.
I worked on customising Microsoft MakeCode for micro:bit for my organisation, and learned something about working with AI tools that I wish I had known on day one.
The problem
I needed to add our own sign-in and backend storage to MakeCode. It sounds simple. It was not.
For 20 to 25 days I threw myself at it in Cursor. The AI generated code, I debugged it, it generated more, I debugged again. The codebase is enormous, and neither I nor the AI could hold enough of it in view at once. We were walking in circles through a maze.
What changed
I stopped asking the AI to solve the problem and changed what I asked for.
- I became the context manager. I stopped asking “how do I implement this?” and started asking “how does MakeCode implement this?”
- I documented first. Step by step, I had the AI help me understand and write down how MakeCode already handles sign-in and project storage.
- Only then did I implement. With that documentation in hand, I switched to Claude Opus 4.5 for the implementation.
What had taken 25 days of frustration was resolved in hours. That was not because one model is better than another. It was because I finally gave it the right context.
AI is a mirror, not a map
AI tools reflect the quality of the context you give them. They are powerful, but they need three things from you:
- A clear, written understanding of the existing system.
- A well-defined problem.
- You as the architect, not the passenger.
Stop asking AI to figure everything out. Use it to deepen your own understanding first.
If you are working in a large open-source codebase: document first, code second. Your future self and your AI assistant will both thank you.