Nothing published yet
Here is what is being written.
What is coming
In the order they go out.
An LLM is not a tool
Three different things get called "tools" in this field — the MCP tools an agent calls, the software a person works in, and the models themselves. Collapsing them is why so much agent tooling is built the wrong way round.
What a memory has to carry to be worth keeping
A sentence is not enough. What we learned writing four kinds of memory, and why the one that took longest — when a practice applies, and what proved it — is the one that changed how agents behave.
Nobody should ever see an agent's view
The packs, the JSON, the face an agent presents to itself: all of it is real, and none of it belongs in front of a person. A note on drawing that line, and what happened the times we got it wrong.
Be told when the first one lands
One email per post.
One email per post. No newsletter, no digest, no "here's what we've been up to".