Page 17
12 articlesThe Latency Budget Your Agent Loop Stole from the Search Box
Swap a 200ms search call for a 4-second agent loop and the latency budget does not vanish — it migrates from infrastructure to UX, and the team that does not catch the handoff ships a worse product with a better metric.
The Library Version Your Coding Agent Remembers Wrong
Coding agents confidently emit code that compiles against the wrong version of your dependencies. The model isn't hallucinating — it's remembering a library that no longer exists.
The Multi-Agent Deadlock That Hangs on Two Calendars
Multi-agent workflows with human approval queues recreate every classical deadlock condition. The cycle hides across two queues and two calendars until a customer notices.
The On-Call Runbook That Assumed a Human Would Read the Page
When an agent posts a polite summary in the incident channel and the commander reads it as ownership, the escalation chain has quietly acquired a transition nobody wrote down. Patterns to close the gap.
The Postmortem Template With No Row for the Model's Inference
Most incident templates have no row for what an AI agent inferred — so the action items chase a deterministic fix for a probabilistic failure, and the same outage class keeps recurring.
The Process Your Agent Quietly Owns Without Documentation
When an agent's prompt becomes the only authoritative description of a business process, you have shipped a runbook in the cloud that nobody can audit, version, or hand off.
The Prompt Cache Your Personalization Layer Quietly Killed
Per-user context at the top of the system prompt is the silent way to triple your inference bill. Here is why the cliff is invisible until billing close, and how to defend the cache boundary in code, in review, and in CI.
The Provider Failover That Multiplied Your Incident Surface
Adding a secondary LLM provider does not make your system redundant. It makes it twice as expensive to maintain — and brittle if you skip the prompt-engineering work that follows.
The Rate Limit You Set for Humans an Agent Saturates in Three Seconds
Rate limits calibrated for human-paced traffic collapse the first time an agent points a planning loop at the endpoint. Treat the limit as a split contract — throughput budget plus abuse ceiling — keyed on tenant and workload class.
The Recurring Task Your Agent Scheduled With Nobody To Inherit
When an agent spins up a recurring task and walks away, the schedule outlives its owner — and the orphan rate compounds quietly until somebody runs the audit.
The Streaming Response That Contradicts Itself
Streaming LLMs surface partial reasoning users read as final answers. Why contradictions inside a single response break UX and evals, and four patterns that put a commit boundary back in.
The Success Metric That Improved Because the Model Declined the Hard Cases
A rising task-completion rate after a model upgrade can mean the agent got better — or that it stopped attempting the hard cases. Decompose your success metric or watch churn pay the bill.
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.
