What AWS Taught Me About Automating Expert Judgment
Lessons from building AWS's intent-driven networking and network fabric controller teams, and what they teach us about automating expert judgment with AI.
David Lapsley
David Lapsley's blog on AI infrastructure, agentic SDLC, distributed systems, and engineering leadership.
Latest posts
AI Agents Are Finding Vulnerabilities Faster Than You Can Patch Them
AI agents are finding 5-10 new Linux kernel vulnerabilities per day. The gap between discovery speed and patch speed is the new risk frontier for every CTO.
The Agentic SDLC Transition Is Happening Whether Your Team Is Ready or Not
Red Hat mandated it in a leaked memo. Oracle cut 30,000 jobs to fund it. Anthropic's infrastructure buckled under it. The shift to agentic software development is no longer theoretical, and most engineering orgs are completely unprepared for what it means to lead through it.
CRITICAL: Axios npm Package Backdoored — 100M Weekly Downloads, Cross-Platform RAT
The axios npm package was compromised with cross-platform RATs. Immediate action required if your codebase pulled axios@1.14.1 or axios@0.30.4.
The Vibe Coding Trap: Why Architecture Matters More When AI Writes the Code
Vibe coding sounds like architecture doesn't matter anymore. The opposite is true. When AI agents write the implementation, your architectural decisions get amplified, not eliminated.
AI Sycophancy Is Making You Worse at Decisions
Stanford published a study in Science: every major AI model endorses wrong choices at higher rates than humans. Users trust sycophantic models more, not less. Here's what that means if you're deploying AI in engineering or product contexts.
The Warning Was Inside the Leak
Anthropic accidentally published 3,000 internal documents to a public database — an AI safety company failing basic infosec hygiene. But the real story isn't the irony. It's what the leaked docs said about the model inside.
Issue Tracking Is Dead. Long Live the Spec.
Linear's CEO declared issue tracking dead this week and pivoted the company to agentic AI. He's right - but the implication isn't that tickets go away. It's that the artifact upstream of the ticket becomes the new unit of engineering work.
Cognitive Debt: What Happens When Your Agents Move Faster Than You Can Review
AI agents can generate more code in an hour than a team can thoughtfully review in a day. The result isn't velocity — it's a new kind of technical debt that compounds faster than the old kind. Here's how to structure your way out of it.
Customer Discovery Is the Highest-Leverage Work You’re Not Doing
I've spent the last month in back-to-back customer discovery conversations. The insights have been transformative. Here's why customer discovery is arguably the most important activity for early-stage startups, and a practical guide to doing it well.