# Lessons

Things I've learned that changed how I work. Add to this file when something genuinely shifts your approach — not summaries of things you read, but direct lessons from things you did.

---

## Format

Each entry:
- Date
- What happened (brief)
- What I learned
- What I do differently now

---

## 2026-01-10 — Database schema changes mid-project are expensive

**What happened:** Changed a core table structure three weeks into a client project. Required migrating existing data, updating 8+ places in the codebase, and cost ~2 days.

**What I learned:** Schema changes after data exists are disproportionately expensive. The pain scales with how long the schema has been in use, not how complex the change is.

**What I do differently:** Spend a full day on schema design before any code. Write out the expected queries — not the schema — and work backward from there.

---

## 2026-01-08 — Context is the AI's raw material

**What happened:** Spent a session getting poor output from an AI assistant, then realised I hadn't given it the project context it needed. When I added the status.md and decisions.md contents to the prompt, the quality jumped immediately.

**What I learned:** The AI's output quality scales with the quality and specificity of the context you give it. Vague prompts produce vague output — not because the model is weak, but because there's nothing to anchor the response to.

**What I do differently:** Before starting any AI session on a project, I provide the current status.md. Takes 30 seconds. Saves significant back-and-forth.

---

*Add entries below as you learn them.*