Page 46
12 articlesAI-Native API Design: Building Backends That Agents Can Actually Use
REST APIs were designed for human-authored clients. AI agents break them in entirely predictable ways — hallucinating endpoint names, retrying without idempotency, ignoring sparse error messages. Here's how to build backends that agents can call reliably.
AI-Native Logging: Capture Decisions, Not Just I/O
Conventional logs tell you what your LLM system did. AI-native logging tells you why — capturing the decision logic, rejected alternatives, and confidence signals that explain production failures.
The AI Onboarding Gap: Why Engineers Can't Learn What They Can't Test
New engineers can't bisect LLM regressions, can't read the implicit constraints baked into prompts, and can't test their way to confidence. Here's the scaffolding that makes AI systems legible to people who didn't build them.
AI as the Permanent Intern: The Role-Task Gap in Enterprise Workflows
AI tools stall in enterprise workflows not because of model quality, but because teams deploy them as if they hold organizational roles they structurally cannot occupy. Here's the gap—and how to design around it.
AI Pipeline Exception Handling: Hallucinations, Refusals, and Format Violations Are First-Class Errors
How to treat hallucinations, refusals, and format violations as first-class error types in production LLM pipelines — with detection strategies and handler patterns for each.
Your AI Product's Dark Energy: The Background Compute Nobody Budgeted
Every AI product with persistent state runs invisible inference that never shows up in your latency dashboards or cost models. Here's how to find it, measure it, and decide whether to kill it.
Why Your Application Logs Can't Reconstruct an AI Decision
Application logs capture execution — not reasoning. AI systems make context-dependent decisions that require prompt versions, retrieved documents, and tool call traces to reconstruct. Here's what separates what SRE teams instrument from what AI compliance actually requires.
Building Trust Recovery Flows: What Happens After Your AI Makes a Visible Mistake
A practitioner's guide to designing trust recovery flows when your AI system makes a visible mistake — covering soft vs hard failures, graceful degradation, undo flows, and the metrics that actually measure whether trust came back.
The Boring AI Manifesto: Why a Single Prompt Outperforms Your Autonomous Agent
80% of AI projects fail while the ones quietly delivering returns are classifiers, routers, and extractors—not autonomous agents. A look at why teams keep building the wrong thing, and a framework for matching AI complexity to actual business value.
Chunking for Agents vs. RAG: Why One Strategy Breaks Both
RAG retrieval and agent execution have opposite chunking requirements. Using one strategy for both silently degrades both. Here's what's actually happening and how to fix it.
Code Ownership Decay: What Happens to Team Knowledge When AI Writes Most Commits
When AI writes most of your team's commits, git blame stops answering the question that actually matters: why. Here's how code ownership decays silently and what engineering teams are doing to stop it.
The Compound Hallucination Problem: How Multi-Stage AI Pipelines Amplify Errors
In multi-stage AI pipelines, hallucinations don't just persist—they multiply. Each stage treats the last output as ground truth, turning a single wrong fact into a confidently wrong final answer. Here's the systems-level problem and how to fix it.
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.
