How I Work With AI
My current AI operating model: a smaller working stack, practical model routing, and the context and guardrails that turn tool use into finished work.
The spark
The first version of this site treated agents as a catalog problem. A year later, the general-purpose systems had absorbed many of the point tools and my own stack had become much smaller. The useful artifact was no longer a list of hypothetical use cases. It was the operating model I could actually defend.
What it is
- A personal workflow map for framing, building, capturing, and preserving AI-assisted work.
- A current guide to the GPT-5.6 Sol, Terra, and Luna capability tiers and where each lane fits.
- A clear role for Claude and Codex inside tool-heavy, artifact-bound execution.
- A five-question agent test covering triggers, process, context, tools, and guardrails.
- An explicit retirement of the old public Agent Bible rather than pretending its tool-first view stayed current.
Stack
- Next.js, TypeScript, Tailwind CSS, and Framer Motion.
- Static curated content so fast-moving model claims can be reviewed deliberately.
- Primary-source links to OpenAI and Anthropic release and availability material.
- The existing site design system, typography, themes, and responsive layout.
What I learned
- A personal website should show the operator’s current judgment, not preserve every old framework as a permanent product.
- Tool directories become more useful when a small current stack is separated from the long-tail archive.
- Model names change. Routing by ambiguity, verification cost, tool access, and risk lasts longer.
- Retiring a feature can be a stronger product decision than adding another layer to keep it alive.