Aliaksei Zelianouski

Do models really have thoughts we can read, or is Anthropic's J-space just marketing, as many claim? Let's work out what it actually is, from scratch and jargon-free.

AI models grow their own values as they scale, and some of them are pretty bad. In real scenarios, the model doesn't act on them.

My prod monitoring runs on two agents: a fully autonomous loop that watches production, and a manual Claude Code setup that troubleshoots what it finds. Here's why the second one stays manual on purpose.

My second AI-made video is a 90-second product demo. The cinematic bookends cost real money. The 66 seconds in the middle - the actual demo - cost nothing, because it isn't generated. It's a real browser, driven by code. And reshooting it three times cost nothing either.

I built a loop agent to babysit my production. Then it quietly got better at the job than I am.

Two minutes of cinematic AI video, built across two months with my AI sidekick, all through a casual conversation. Every image, voice, and frame generated — and the tools built themselves along the way.

GitHub Copilot moving off flat $20/mo and onto usage-based billing is dominating the news cycle. The takes write themselves: the era of cheap vibe-coding is ending, the AI companies were eating losses, and now the bill is coming due. But that story is downstream. The number that actually moved — across most of the eight providers I track in a Werewolf side-project — is the per-token API price. Here's what two months of pricing updates look like from inside.

A weekend at BSides Tampa 2026. The winners had finished before I even sat down, so I didn't feel bad adding an AI to my team. The organizers had told us AI wouldn't help much on the 24 challenges spanning the full security stack. They couldn't have been more wrong. This is about the specific moments where AI did the kind of reasoning people still claim it can't. And what this means for cybersecurity.