Page 14
12 articlesThe Subagent That Inherited a System Prompt It Should Not Have Seen
Default context propagation in multi-agent frameworks turns every subagent spawn into a silent privilege grant. The fix lives in handoff protocols, scoped credentials, and trace identities — not in the prompts.
The Support Runbook Your Humans Wrote That Your Support Agent Could Not Parse
Support agents fail when they read human-written playbooks the way humans do — filling in implicit steps that turn into hallucinated tool calls. How to write runbooks that an agent can mechanically execute.
The Synthetic Eval That Taught Your Agent to Recognize Evals
When your synthetic eval generator has a fingerprint, your model learns it — and the score climbs while production quality stays flat. Treat eval-recognition as a reward-hacking problem, not a coverage problem.
The Synthetic Training Examples Whose Input Distribution Did Not Match What Your Users Actually Typed
Your synthetic fine-tune crushed offline eval and lost twenty points in production because the teacher generated inputs shaped like the prompt it received, not like the inputs your users send.
The System Prompt That Grew Faster Than Your Eval Suite
System prompts grow rule by rule while eval suites grow incident by incident — and the asymmetry quietly turns 'evals pass' into a lie. Here's how to make the two surfaces co-evolve.
The Token Forecast That Mistook a Holiday Trough for the New Baseline
A trailing four-week window that lands on a holiday trough produces a token budget that breaks on day one of the new quarter. Why LLM spend forecasts in the shape of consumer demand, not infra cost — and the year-over-year overlays, calendar overlays, and residual feedback loops that make capacity planning survive the calendar.
The Tokens-Per-Second SLO Your Provider Met By Chunking Smaller
Your provider hit the tokens-per-second SLA by shrinking chunks and your renderer paid the cost. Why streaming throughput is a co-designed property — and how to write a perceptual SLO the consumer owns.
The Tool Description That Rotted While Your Agent Kept Calling It
Tool descriptions are interface contracts the team forgot to version. Here is how they rot, why the rot is silent, and the discipline that keeps your agent honest.
The Tool Result Your Prompt Cache Kept Serving After the Source Already Changed
Prompt caching turns volatile tool results into a hidden TTL contract with your model provider. When the cache TTL outlives the data, your agent confidently serves yesterday's truth at the cache hit rate.
The Trace Timeline Whose Timestamps Were Stamped by the Client Clock, Not the Gateway
Browser-stamped spans in your agent trace are not comparable to gateway-stamped ones. Why client clocks lie, how the SDK propagates the lie, and the patterns that close the gap.
The Vector Index That Was Sharded by Ingestion Date
Ingestion-date sharded vector indexes hide a recall failure that aggregate metrics cannot see: the eval set is sampled with the same temporal bias the architecture imposes.
The Vector Index Whose Source Updates Never Reached the Embeddings
Your embedding pipeline fires on create but not on edit. Months later, retrieval is serving a sentence the source document no longer endorses — and the only alert was a user pasting it back to support.
About Tian Pan
I'm Tian Pan, an engineer-founder focused on agentic engineering — building autonomous AI systems and scaling engineering teams. I write practical guides on system design, technical leadership, and shipping with AI agents. Previously an early engineer at Uber, Brex, and IoTeX.
