#advanced
Every post tagged "advanced".
48 posts · showing 21–30
-
The Failure Modes That Don't Improve With Scale
Most agent limitations improve with the models. A smaller set won't, because the constraint isn't the model — and building elaborate workarounds for one while waiting out the other gets both wrong.
-
Upgrading One Agent in a System of Six
A contained improvement to one classifier made the supervisor above it drop half its escalations. No error, no failing test — because each agent's own tests passed against its own version.
-
Half the Job Is a Real Answer: Designing for Partial Success
Thirty-one of forty invoices reconciled. Reporting success hides nine; reporting failure throws away thirty-one. The middle is where most runs land, and most designs can't express it.
-
Which Agent Alerts Should Wake Someone Up?
An agent can be green on every service metric while doing the wrong work at ten times the cost. One rule sorts the alerts — and the highest-value one is the one almost nobody has configured.
-
The Tool That Changed Under You: Supply-Chain Risk in Agent Tooling
No deploy, no prompt edit, and an ops agent started restarting services mid-investigation. A maintainer had improved a tool's documentation — and documentation, in an agent, is instruction.
-
One Query Is Rarely Enough: Iterative Retrieval for Agents
The search wasn't broken — it answered the question as typed, which isn't the question that was asked. Making retrieval a tool the agent calls repeatedly needs one signal most retrievers refuse to give.
-
Computer Use: When the Agent Drives a Whole Desktop
Clicking is easy; knowing what to click is a perception problem solved from a screenshot. The failures are modals, focus loss, and scroll state — and an agent that assumes its click landed compounds errors silently.
-
Request, Worker, or Durable Execution? Hosting Models for Agents
Where a run lives decides how long it can be, whether it survives a deploy, and whether it can wait for a person. Pick by the longest wait it can hit — and keep state out of process memory so the answer can change.
-
Red-Teaming Your Own Agent: A Practical Injection Suite
Trying a few "ignore your instructions" prompts tests the easiest attack against the strongest defense. A real suite is a matrix of carriers, payloads, and targets — asserting on actions taken, not words said.
-
Shared Tool Access: When Two Agents Touch the Same Resource
An agent holds a stale read for seconds, not microseconds — long enough that last-write-wins silently deletes another agent's work while both report success. Partition, append, or version.