Tag: AI
All posts tagged “AI”.
-
Why Trusq runs open-weight models, even in the cloud
Every model that writes, classifies or checks text in Trusq has open weights, and for ordinary workspaces Mistral runs them in its EU cloud. Whether a model is open decides what we can do with it. Who runs it decides who sees the data.
-
No code makes my AI reviewers take turns
In claude-agentic-kit, two model families take turns reviewing every specification and plan, a cheaper model applies the fixes, and a script decides when the loop is done. When I checked which of these rules the code actually enforces, the switch between the two families turned out to be only an instruction in a prompt.
-
I measured my framework on features we had already shipped
To test a change to claude-agentic-kit, I regenerated the plans for four Trusq features that were already in production and compared them with what we had actually built. Fragile line references fell from 64 to 0, and one feature built end to end avoided two issues a human reviewer had caught.
-
Who judges the judge?
At Trusq a second model checks every AI answer before it is sent. When we compared models for both jobs, the judge mattered more than the writer.
-
I compared my agent framework with Spec Kit, Superpowers and the rest. Here is what is actually different
Spec Kit, Superpowers, Matt Pocock's skills, BMAD, OpenSpec, GSD Core and Kiro, checked against their own documentation: what is actually different, and what the others do better.
-
Scripts decide, models write
A coding agent gives a different answer every time you ask. How claude-agentic-kit puts a deterministic frame around that non-deterministic process, so every decision that steers the agents is made in code you can test.
-
The model that thinks it's Claude
Kimi K3 sometimes claims to be Claude. Washington calls it theft; critics call it politics. Both miss the point: the distillation fight decides the price of AI.