All posts
Everything published here, newest first.
316 posts · showing 61–70
2026
10 posts-
Your Retriever Returned 12 Chunks. The Agent Only Really Read 3 of Them.
The correct passage was in the prompt and the agent wrote around it anyway. That's not a retrieval bug — it's a placement bug, and there's a one-minute test that tells the two apart before you touch your embeddings.
-
Blackboard or Mailbox: How Should Three Agents Share What They Learn?
Two agents discover the same fact; a third contradicts both and nobody notices. The shape of your shared state — a common board or point-to-point messages — decides which of those happens.
-
Your Agent Has One manage_calendar Tool. That's Why It Deleted the Wrong Meeting.
A single do-everything tool looks tidy in the schema and hides four separate problems: guessed actions, unvalidated payloads, permissions that can't tell reading from deleting, and traces that flatten three risk levels into one name. Here's where to cut instead — and where splitting makes things worse.
-
Your Agents Asked 40 Questions This Week. Which Ones Should Have Reached a Human?
Human attention is the one resource in an agent deployment that never shows up on the bill — so agents spend it freely. Here's a three-way routing rule, a threshold you can actually tune, and the two metrics that tell you when your agent is asking too much or too little.
-
Map-Reduce or Sequential Refine? How an Agent Should Read 400 Pages
Chunking 400 pages of contracts forces a choice most teams make by accident. One shape quietly drops evidence as it goes; the other can never spot a conflict between two documents — and the fix is not the pattern, it's what the map step returns.
-
Draining a Data Backlog: Should the Agent Batch Rows or Take Them One at a Time?
The ratio of preamble tokens to row tokens decides whether chunking a backlog saves an order of magnitude or nothing at all — and it's the alignment, retry amplification and blast radius that decide whether you should anyway.
-
Do You Need an Agent Framework? Write the 40-Line Loop First
The agent loop is shorter than the quickstart for the framework that hides it. Write it once — then you can tell which framework features you actually need and which ones you'd be adopting on faith.
-
Green Evals, Broken Production: The Frozen Fixture Problem
An eval suite built on recorded tool responses stops testing your agent and starts testing a museum. Here are three cheap detectors for fixture rot — and the rule for when a case should call the live tool instead.
-
Who Decides the Agent Is Done? Three Stop Conditions, and Why "The Model Said So" Isn't One
Most first agents end their loop when the model stops calling tools — which means the model, not you, decides what "done" means. Here are the three stop conditions a real loop needs, and the twenty lines that turn a silent wrong answer into another iteration.
-
Read, Draft, Send: Where to Draw the Permission Line in a Personal Email Agent
Most email agents get permissions wrong by granting them per tool. The safer line runs through the arguments — who the recipient is, how many there are, and whether the thread was already open.