Page 34
12 articlesWall-Clock Deadline Drift: Why Your Agent Thinks It Has Time It Doesn't
Transparent tool retries silently burn wall-clock budget while the planner reasons against a stale deadline, producing a bimodal SLA failure that no single layer's metric catches.
The Curious Customer: Designing AI for Users Who Treat Your Agent as a Puzzle
A third population sits between cooperative users and malicious attackers: the curious customer who treats your AI agent as a puzzle. Here is how to build evals, refusals, and fallbacks that hold up when they are the moment your brand is being judged.
Capacity Math for Agent Loops: Why Your Provisioned Throughput Is Half of What You Think
Provisioned throughput sized on user QPS quietly under-provisions agent products by the loop fan-out factor. Plan with model-call rate, loop depth, and burst tails instead.
Agent State Diff: Why Eyeballing Two Traces Doesn't Scale
Two agent runs of the same prompt almost never produce the same output. Diffing them at the text level hides the actual cause. Here is what structural diffing requires and how to build for it.
The Onboarding Gap: Why New Engineers Take Three Months to Touch the AI Stack
AI codebases carry a hidden domain-knowledge tax that turns three-week ramps into three-month ones. The fix is decision history, not architecture diagrams.
The AI Wallet: Why Token Budgets Belong in the UI, Not the Engineering Dashboard
Hiding AI cost from users ships silent throttling and surprise downgrades. Treating the token budget as a real product surface — preview, caps, model selection — turns the cost ceiling from a churn driver into a monetization lever.
The Annotator Calibration Gap: When Human Raters Quietly Stop Agreeing
Six months into a human eval program, the inter-rater agreement number is a weighted average of three different implicit rubrics. The model didn't drift — the measurement instrument did.
The Audit Trail Mismatch: When User, Agent, and Tool Each Have Different Logs
Agent stacks emit four logs that don't agree. The fix isn't more logging — it's a transaction ID minted at the user-action boundary, a unified audit record, and retention sized to the compliance question, not the subsystem.
The Bypass Vocabulary: When Users Learn to Jailbreak in Polite English
Production AI products leak refusals to three-word framings — 'hypothetically,' 'for educational purposes,' 'for a story.' How to detect and defend against the bypass vocabulary your users pick up from social platforms.
Compliance Reviewer as Eval Author: Why Legal Should Be Writing Your Test Cases
Compliance reviewers spot LLM failure modes engineering evals systematically miss. Move them out of document-review gates and into the regression suite — legal sign-off becomes a statement about pinned test cases that run every commit.
Context Bloat: The AI Memory Leak You Cannot Grep For
Long-running agent sessions silently leak tokens — quadratic cost growth and quality degradation hide inside conversation history. How to instrument, prune, and compact it.
Conversational REST: When Your Chat UI Needs Pagination, Filters, and Sort
Chat is a great input modality and a terrible output one. The moment your agent returns more than three results, the right answer is to render UI — not to keep talking.
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.
