How we think. What we make.
Build logs, system breakdowns, and essays on private AI, local-first architecture, autonomous agent systems, and the philosophy behind sovereign intelligence.
Latest
The hardest product decision in an AI system isn't what model to use. It's how to make one engine serve operators who do completely different work — without f...
A vector database, an MCP server, and a self-improving agent walk into a loop. What emerges isn't a pipeline — it's compound interest on machine intelligence.
The hardest design decision in an autonomous system isn't what the agent can do. It's when the agent should stop doing it and let a human take over.
An agent that's 80% accurate isn't 80% useful. The real math involves rework costs, supervision overhead, and the break-even threshold that determines whether y...
The final layer. An agent that can be trusted to operate autonomously needs to know when it's right, when it's wrong, and when to ask for help. That calibration...
Without governance, autonomous systems reliably drift toward waste. With it, they stay focused and predictable. Here's how declarative rules replace hardcoded l...
Most agents start every session from scratch. That's not a limitation of the technology — it's a limitation of how the systems are built. Persistent semantic ...
Generating code is easy. Generating code that compiles, doesn't break existing tests, and actually solves the problem — that's where the real engineering live...
Most agent failures aren't execution failures. They're planning failures — good work on the wrong task, in the wrong order, at the wrong level of detail. The ...
Before an agent can reason, plan, or generate, it has to see what it's working with. Most agents still read code the way a text editor does. That's the root of ...
A series announcement. After months of building and running autonomous AI systems, we've identified six layers that every serious coding agent needs. Each one g...
We built a research system that runs around the clock on local infrastructure. Here's what changed when we gave it a real embedding model and a vector database ...
What happens when you rebuild the AI inference stack from scratch in Go — no Python, no ONNX, no C dependencies. A look at a novel approach to model compilati...
Autonomous agents need more than instructions — they need governance. Here's how we built a declarative rules engine to keep our AI systems trustworthy.
Why we run our own stack — and what it actually costs. A look inside the sovereign AI infrastructure powering a small creative studio.
The case for local-first AI: sovereignty over your data, independence from platforms, and systems that compound instead of expire.
How we're using ChatGPT for stills, Grok for animation, and Codex for production management to build a full animated series — with a team of zero animators.
How we built a 10-process autonomous AI system with cost breakers, structural code analysis, and self-correcting code generation — running locally on a Mac Mi...