Guides
How to build agents that survive production
Everything here comes from watching a lot of agent runs fail in interesting ways. No prompt tricks — just the loop, the tools, the tests, and the bill.
Architecture
9 min read
Designing an agent loop that survives production
Almost every agent that dies in production dies the same way. The fix is not a smarter model — it is a loop built from steps that can each fail independently.
Mara Vane
14 January 2026
The unit of work is the step, not the run.
Tools
Tools that fail well
The model will pass your tool something absurd — reliably, on a long enough timeline. Whether that becomes a corrected second attempt or a support ticket is entirely down to how the tool is written.
7 min read
Evaluation
How to actually evaluate an agent
Most agent evaluation is a number that goes up. When it goes down you have no idea which of the eleven steps got worse — here is the boring, specific alternative.
8 min read
Cost
What a run really costs
The bill is never the model. It is the retry you did not notice, the context you resent forty times, and the planning step running a frontier model to pick between three tools.
6 min read
Or skip the reading and run one
Every idea in these guides is already built into the runtime. Install it and the defaults are the advice.
Apache 2.0 · No account needed to start