Notes from the workshop.
Product updates, engineering notes, and thinking on privacy-first AI tooling.
AI for indie musicians: a 2026 toolkit
AI is a session musician you can hire at 2am for a dollar. Four pillars that earn their keep, and the parts of being an artist no model belongs near.
The last three months in AI: Audio models
Realtime voice latency fell under 200ms in four weeks. Music spent the quarter in court. Five sub-modalities, five clocks.
The last three months in AI: Video models
OpenAI killed Sora. The Chinese open-weight stack took the top three on the arena. And a 10-second clip now starts at fifty cents.
The last three months in AI: Text models
Six flagships in eight weeks, an open-weight model that tied GPT-5.5 on coding, and Meta walked away from Llama. The Feb–May 2026 LLM catalog.
The last three months in AI: Image models
Reasoning came to image generation, native typography stopped being a tell, and the leaderboard turned over twice. The Feb–May 2026 catalog.
Most AI work can wait. Your bill shouldn't.
Every major AI vendor charges half price if you'll wait a few hours. Most teams pay full price for work nobody is waiting on. A workload-sorting frame, and the four vendors' fine print.
Serious software is local. AI is finally catching up.
Photoshop is local. Excel is local. Your IDE is local. AI was the one category that wasn't, and the technical excuse for that just expired.
You don't need every AI model. You need the right five.
2.88 million models on Hugging Face. 359 on the leaderboard. You can do almost any job with five. The case for curation over catalogs.
AI products are mortal. Your workflow shouldn't be.
Eight obituaries in 24 months, nine Claude models retired, a four-hour ChatGPT outage. The case for keeping your daily AI workflow on a machine you own.
Regulations are eating cloud AI
Eight jurisdictions, billions in fines, and the moment 'send it to a cloud LLM' became the riskiest default in regulated work. A compliance map for the people drawing the AI architecture.
Cost-per-task, not cost-per-token
Your AI bill probably has the wrong column header on it. Six realistic tasks, six providers, and a table that re-ranks the leaderboard once you stop measuring in bytes and start measuring in outcomes.
Choosing a local model in 2026: a flowchart
Roughly forty open-weight models worth installing. You need one. Tell me what you want to do and what laptop you have. Here's a five-by-five grid with the pick in each square.
AI for students who don't want to cheat
AI is the best private tutor a student has ever had, if you use it as a tutor. The line professors actually care about, and how to stay on the right side of it without giving up the most useful study tool of your life.
Offline AI is more useful than you think
On a plane. In a hospital handover. In a kid's bedroom. A field-tested case for AI that runs on your machine, beyond the privacy lecture.
Subscriptions ate my month, and AI shouldn't join the pile
Three trial-ending emails this week. Forty-one percent of households say they feel it. AI is about to become the thing you reach for every day. Does it really belong on auto-renew?