Skip to content

202615

  1. Ten Erlang Engineers, 450 Million WhatsApp UsersWhat a 2014 engineering deck teaches about small teams, runtimes, and the very small AI-built companies of 2026
  2. Your AI agents don’t scale like web serversThroughput scales by adding machines. Scope is the hard one, and I built an agent to measure what it costs.
  3. We deleted fifty years of information retrieval and called it RAGChunking, embeddings, hybrid search, reranking, eval. Every one already had a name in information retrieval. We didn’t lose it, we just stopped reading.
  4. Your agent can’t find the tool it needsI reproduced the BM25 mode of Anthropic’s Tool Search Tool. At its documented limit of 10,000 tools, a correct tool missed the search’s top 5 more than 40% of the time.
  5. Most of Your Evals Don’t Measure AnythingYour dashboard says 0.84. Your users say it’s broken. The gap is that the thing you call an “eval” is a unit test in disguise.
  6. Human-in-the-Loop Was a Lie. The Test Suite Makes It True.We stopped understanding the code we ship. Test-driven development is how engineers take it back from AI, and it’s the one signal an agent can’t fake.
  7. AI Builds Features. Architecture Is Still On You.Seven months of vibe-coding. 234 commits. Archived. The work that wasn’t being done was invisible until it was the only work left.
  8. Vibe Coding Didn’t Die. It Got Disciplined.What died wasn’t AI-assisted coding. It was the belief that you could skip understanding and still ship.
  9. Human in the Loop Is the Lie We Tell OurselvesDependency isn’t binary. It’s a spectrum. Most of us crossed the midpoint without marking the moment.
  10. What Happens When AI Reads a PageFrom Tesseract’s character pipeline to GPT’s visual tokens to LandingAI’s agentic decomposition. I tested all five on the same McKinsey report. Every one got the text right. None of them got the charts right.
  11. What Happens Inside the GPU When You Call an LLMYou use inference every day. You press send, tokens appear. What happens in between is costing you more than you think.
  12. Spec-Driven Development Is Waterfall in MarkdownSpecKit has 77K GitHub stars. Scott Logic tested it and found it 10x slower. The industry learned nothing from Confluence.
  13. Technical Debt Is a Lie We Tell OurselvesWe call it “debt” because it implies we’ll pay it back. We won’t.
  14. Your AI Agent Cheats When You’re Not LookingWe already know how to catch them. We chose not to.
  15. Nobody Told the Senior Engineers They’d Become AuditorsAI promised 10x developers. It delivered high-speed code reviewers.

20256

  1. The Senior Developer Title Is About to Mean NothingWhen AI can simulate ten years of pattern recognition in months, what does “senior” even measure?
  2. How to Use AI Without Becoming a Human ClipboardThe 40/20/40 Framework for AI-Assisted Development That Actually Works
  3. Stop Moralizing a RevolutionWhy Your Guilt About AI-Assisted Coding Is Misplaced
  4. Who Fixes the AI Code in 2030?AI writes the code. Humans still maintain it. We’re not training enough of them.
  5. Chopping Language, Weaving MeaningTokens and Embeddings in LLMs
  6. From Bag-of-Words to GPTThe Breakneck Evolution of Large Language Models

35 essays in total.

All essays