Practical notes from backend engineering and building products, plus occasional writing about business, family, and ambition.
19 notes for “AI”
I started with Grok Bot and ended up looking at the infrastructure underneath every agent that is allowed to do real work. Daytona is one of the more interesting bets, but this market will be won on trust, not demos.
What x402, Coinbase, Cloudflare, and always-on agents such as Grok Bot reveal about the next payment layer of the internet - and where Bitcoin fits.
Notes from my current Tilt workflow: why I run bounded overnight work locally on a MacBook, how the tools fit together, and where human review remains.
How I organized Codex into twelve bounded product workstreams without turning one shared repository into chaos.
A personal framework for finding durable upside in technology, from steam and electricity to language models, AI agents, Bitcoin, and whatever comes next.
What started as a quiet plan for planting trees became an interactive 2D and 3D masterplan for 8,615 square metres of family land.
A technical manual for running a one-person internet company on Hetzner with Cloudflare, Caddy, PostgreSQL, Redis, systemd, GitHub Actions CI/CD, R2 backups, mobile Codex operations, and resource limits.
What started as a personal memory app is becoming a context layer for people and AI agents, without abandoning the personal product that exposed the real problem.
Giving an AI assistant more text is easy. Giving it current, permitted, traceable, and testable context is an engineering problem.
Why AI-assisted development gets safer when agents can read the operational runbooks before they make code or deployment decisions.
Why agent memory is only useful when every durable fact points back to a source, a date, and a reason it should still be trusted.
A practical structure for giving coding agents useful context without turning every task into a giant prompt dump.
For PLAYGRND, WhatsApp is not a gimmick. It is the channel users already trust, so login and profile claims should meet them there.
PLAYGRND is my practical example of how a small team can move faster with AI without letting AI replace product and engineering judgment.
Agent reliability is not only retrieval. The hard part is deciding which source is allowed to control the next action when context conflicts.
Why agent memory is useful, but not enough. The real work is source-backed context, permissions, stale-context detection, and eval traces.
Why the long-term advantage in AI-assisted software work is not picking the winning model, but building reliable context systems around any model or tool.
How I set up mobile first development and direct production deploys using an old laptop, Tailscale, tmux, Telegram notifications, Cloudflare Tunnel, and Wasp.
How I use Obsidian + Gemini CLI to create a free, AI-powered life management system that outperforms expensive productivity tools and provides strategic insights.