blog

For two weeks GPT-5.6 was the most impressive model I had ever used. Then Claude Fable came out. Second place has never been this good, and it has never mattered this little.

The demos are simpler than they look. Give Fable the goal instead of the steps, fence it with house rules, set a bar for done it can't talk its way out of, and loop it until it gets there.

A practical framework for briefing AI agents well: give them the right context, clear constraints, and the exact shape of the output you want.

GPT-5.5 is stronger almost everywhere, but the strangest thing is that the upgrade often feels subtle because frontier coding models are already so good.

A personal note for non-tech friends and family on what AI is starting to change.

The first coding model I can start, walk away from for hours, and come back to fully working software. Judgment under ambiguity + strong validation changes everything.

Why I rely on GPT-5.2 Pro: the slow, long-thinking mode, when it's worth the wait, how to prompt it, where it fails, and whether $200/month pencils out.

Two-week hands-on: better instruction-following and codegen, Pro is a slow genius, but standard is too slow for daily use.

Slow, deliberate GPT-5.1 Pro vs Gemini 3: when I escalate for backend/infra and deep research, why the UX still hurts, and why this thing is fucking scary smart—arguably a better reasoner than most humans.

Hands-on Gemini 3 review: A look at performance, capabilities, and how it feels to use compared to GPT-5.1.

I thought GPT‑5 was incremental—until it shipped a complex prototype from a spec in one hour. This piece explains why speed + reliability make it feel like real software, the tradeoffs to watch, and how to use the right modes to get the most out of it.