All posts
Everything published here, newest first.
316 posts · showing 91–100
2026
10 posts-
Three Ways to Get Valid JSON, and What Each One Costs
The retry loop was working and quietly biased: the tickets needing three attempts weren't a random sample, and their eventual valid output was thinner than the truth.
-
Test the Tool or Test the Agent? What Each Layer Catches
Sort a month of eval failures by root cause and most turn out to be tool contract bugs — findable in milliseconds by a unit test, currently being caught by the slowest instrument you own.
-
Choosing Your First Three Tools
Find, read, finish. The third one is the one people leave out, and it's the one that turns "the agent stopped talking" into something you can actually check.
-
Injection That Crosses a Handoff
The fetcher had no write tools and the analyst never saw a web page. The separation held perfectly — and routed the payload around itself, through the one field that carries prose.
-
Your Trace Store Will Outgrow Your Model Bill
Storage passed inference cost in three weeks, and the proposed fix — sample at 5% — would have discarded exactly the runs worth keeping. A trace isn't one artifact with one lifetime.
-
When a Worker Returns Garbage: Validating at the Handoff
Thirty-nine extracts were fine. The fortieth had every field present, correctly typed, and null — and the supervisor aggregated it, because well-formed looked like correct.
-
The Timeout That Looked Like an Empty Result
The health check timed out, the handler returned an empty list, and the agent reported all clear while two services were down. An empty result and a failed check must not share a shape.
-
Budgeting the Context Window: Dividing a Fixed Space
The window wasn't full and the constraints still vanished — because six claimants compete for one space and nobody allocated it. Making history the residual rather than a claimant is what makes the scheme work.
-
What Ten Years of Agent Tooling Might Look Like
Specific capability predictions age badly; layer predictions age better. Which parts commoditize, which stay differentiated — and the one prediction here most likely to be wrong.
-
The Skills That Were Always the Job
Almost nothing on the durable list is new. These are what experienced engineers always said mattered — the economics just didn't reward them proportionately until now.