Skip to content

20266

  1. 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.
  2. 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.
  3. 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.
  4. 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.
  5. 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.
  6. 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.

20251

  1. From Bag-of-Words to GPTThe Breakneck Evolution of Large Language Models

35 essays in total.

All essays