Starting the Lab
The question
What does an experienced software architect need to learn to build useful, reliable things with AI?
How this lab works
Every experiment is one page with the same shape:
- The question — what I'm trying to find out, in one sentence.
- Setup — models, tools, data, and code (linked when it's public).
- Log — dated notes as I go, including the dead ends.
- Results — what actually happened, with numbers where possible.
- What I learned — especially through an architecture lens: cost, latency, failure modes, operability.
Each experiment carries a status: idea, in progress, shipped, or shelved.
First experiments on the list
- Call an LLM API from scratch and understand tokens, context windows, and cost
- Build a small retrieval-augmented (RAG) Q&A over my own case studies
- Write evals before tuning prompts — how do you test a non-deterministic system?
- Give a model tools and build a small agent; map where it fails
- Sketch a reference architecture for an AI feature inside an existing enterprise system
Log
- 2026-09-25 — Site is live. Lab is open.