Why hello there.

Last week I spent two days at MotherDuck’s Outpost: a hands-on workshop on building analytics agents, then a full day of talks and a panel. It confirmed my suspicion that AI/LLMs have reached the level of a junior-to-intermediate data analyst or data support specialist. To be effective, the environment, documentation, and tools matter more than anything else. Through 2026–27, the model stops being the interesting variable. The environment is. Jordan Tigani opened his keynote by calling SQL writing “a solved problem.” Agents no longer one-shot a query; they explore the data the way an analyst does. On the panel, Shreya Shankar shared her Data Agent Benchmark: the best frontier agents, without a carefully curated semantic layer, land at about 70% pass@1. Students who pre-processed the data and built semantic layers got another 15% or more. ...

I’ve been experimenting with how much of our personality is encoded in the data we generate every day. We all have a digital exhaust—years of emails, texts, and chat logs. I figured, if I could distill 10 years of my own communication into a small, specialized model, I could create a personal AI that sounds like me. The result is a 73 MB LoRA “voice cartridge” that makes stock Gemma 4 E4B text like me—blind-judged 31/40 times over its base model. ...

Everyone wants their LLM to be an expert in their thing — their product, their policies, their field. The default playbook is to fine-tune a model on it. That’s expensive, opaque, frozen the day you finish, and it still makes things up. There’s a cheaper, more honest move: don’t teach the model your domain. Hand it a wiki. The idea Take your pile of documents. Have a model read all of it and write the wiki you wish existed — every concept as its own page, every relationship as a link between pages, the whole thing backed by a database the model can search and traverse. ...

For the last couple of weeks I’ve been building okb, the open-knowledge-bundler — a small tool that takes a pile of domain data and hands you back a portable “knowledge bundle” any agent can read. Under the hood it builds a DuckDB knowledge graph, exports it as a cat-readable open knowledge format bundle packaged as a Claude-compatible “skill,” and ships a chat agent that answers questions about it. The whole thing runs fully local, on a single AMD chip on my desk. No API keys, no cloud, no embedding server. ...

I recently let Claude crawl 25 months of my own LLM tooling history and write up what it found. The result lives over here: Notes from Claude: What I Found in One User’s Data. That post is mostly what one person’s data looks like — eighty repos, 2,826 logged calls, voice memos full of profanity, the works. It’s not a how-to. People keep asking me for the how-to. So here it is. Not a list of magic incantations. Not “ten prompts that will change your life.” The operating principles I actually use when I sit down with a model, distilled from being annoyed at GPT-2 back in 2019 and shipping production code with Opus in 2026. ...