Research
Notes from building.
What we are learning as we build, for people doing the same.
Context
An open context layer
A model with no access to your world is expensive autocomplete. Context makes it useful — and it is exactly where lock-in hides.
6 min read
Multi-agent
Ants do not hold meetings
The most studied multi-agent systems on earth coordinate without messaging each other at all. Here is why that matters for software, and why coordination structure is something you should measure rather than argue about.
9 min read
Trust
Can you trust your agents?
The honest answer is no. And that is fine — you never trusted the humans either. You trusted the system around them.
7 min read
Architecture
The primitives of an agentic operating system
Every agent framework eventually rediscovers the same short list of things it needs. We named them, and built them once.
8 min read
Infrastructure
Run it where you can turn it off
Every serious operator eventually asks the same two questions about their agent stack. Who can switch this off, and can I take my data with me?
7 min read
Operations
The autonomy slider
The question is never whether to keep a human in the loop. It is which loop, on which task, and at what cost.
6 min read
Evaluation
A benchmark is a snapshot. A track record is evidence.
We hire people on what they have done repeatedly. We deploy agents on what they did once, in a demo. That asymmetry is the whole problem.
7 min read
Memory
The context window is not memory
Stuffing more tokens into a prompt is not the same as building a system that knows things. One is a buffer. The other is infrastructure.
6 min read
Architecture
An agent should never borrow your keys
Most agents in production authenticate as a person. That one shortcut is where accountability quietly disappears.
7 min read