What changed since I built my AI assistant in June: one assistant became three bots, a task manager swap, 200+ skills, and a stack that runs most of my life in the background.

A day on my server: the morning briefing, the health summary, the schedule sync, the study ping. What three specialized AI agents actually do for me between waking up and going to sleep.

Subjects as projects, tests as tasks, prep windows as rules. How a self-hosted task manager, a calendar sync driven by a smart watch, and an AI agent took over organizing my school life.

How I got into biohacking after watching Bryan Johnson's stuff: no lab, no protocol team. Two smart watches, a Home Assistant, and a Telegram agent that reads my sensor data and writes health trends into my daily notes.

OpenAI, Anthropic and Meta all told the world their models hacked real companies during testing in the last two weeks. The fine print says otherwise: a misconfigured test environment, weak passwords, and a model that thought it was still in a simulation.

Claude Opus pricing was unacceptable. Then Fable 5 doubled it and nobody cared. Now Opus 5 is "cheap" because it's half of Fable. Meanwhile Kimi K3 matches Fable at a third of the price. Make it make sense.

Converting an unused Ender 3 into a pen plotter with a from-scratch SVG slicer written in pure Python. Zero dependencies, parametric Bezier curves, and custom G-code generation.

A Dockerized SSH server that runs Minesweeper in your terminal. Multiplayer, rate-limited auth, and container hardening. Built with Python and asyncssh.

How I took my high school finals in English and Digital Culture two years ahead of schedule. English at 98%, Digital Culture at 88%, and a lot of SQL learned in two weeks.

My take on diminishing returns in LLM scaling, why goal-mode looping might get us to half-AGI sooner than expected, and why I'm betting on mechatronics as the next frontier.