What I Learned
Building with AI.
Critical thinking, real projects, and the theories that changed how I work — written as I go.
7
Deep-dive articles
25+
Lessons documented
100%
From production
Deep Dives &
Critical Thinking
Long-form thinking on the ideas, patterns, and hard problems that shape how I build with AI.
The Forge: Knowledge That Compounds
AI agents are powerful but forgetful. The Forge gives them a persistent knowledge graph — so every correction, pattern, and hard-won fix compounds across sessions instead of evaporating.
Recent Thinking
From foundational patterns to production war stories, each post is a lesson earned the hard way.
The Prompt Is Not the Product
Everyone obsesses over prompt engineering. But the real engineering happens in the systems around the prompt — validation, fallbacks, and graceful degradation.
Evals Are Unit Tests for Vibes
Traditional testing breaks down with LLMs. Evals are our best replacement, but they come with trade-offs nobody warns you about.
Context Windows Are a Lie (Sort Of)
128k tokens sounds like a lot. In practice, the usable context is much smaller — and how you fill it matters more than how much you have.
Ship the Feature Without the AI First
The best AI features I've built started as manual processes. Here's why building the non-AI version first leads to better AI products.
The Temperature Misconception
Most developers treat temperature as a creativity dial. It's actually a confidence calibration — and misunderstanding it leads to subtle, hard-to-debug issues.
RAG Is Not a Search Engine
Retrieval-augmented generation looks simple in diagrams. In practice, the retrieval part is where most teams get stuck — and the failure modes are subtle.
28 Days
Later...
Real usage data from building with Claude Code every day — 28 days, 67+ messages a day, across a full-stack codebase.
1,893
Messages sent
168
Sessions
52,981
Lines written
418
Files touched
What Worked
Full-Stack at Scale
Complex features driven end-to-end — AI provider CRUD, insight pipelines, call centre platforms. 40 successful multi-file changes across TypeScript and Go.
Rapid Bug Triage
41 bug fix sessions with a fast first-responder pattern. TypeErrors, null refs, race conditions — diagnosed and shipped in minutes.
Iterative UI Vision
Pushing through multiple rounds of visual feedback until the result matches the mental model. Dark mode, component layouts, data viz — all polished to spec.
What Helped Most
Languages
Where the Time Went
~45
Lead Gen Platform
~20
AI Features & Playground
~25
UI/UX Polish
~12
Call Centre & Admin
~8
Strategic Planning
Key Pattern
“A fast-iterating visual perfectionist who gives high-level direction, lets Claude execute broadly, then aggressively corrects course when the output diverges from the mental model.”