The 3 AM Nightmare Last week, I let an AI agent run loose on my production server. It was fine — until 3 AM. To interact with the agent, a user must first authenticate across Gmail, a support desk, and a payment platform — all before the agent takes its first action. Permission denied. Permission denied. Permission denied. Three different connectors. Three different auth systems. One very tired
Background A nasty surprise Last summer while trying to deliver a feature for one of our customers, I encountered a nasty situation. The software we were developing, depended on a production grade license of Gurobi. People were on vacations except of my team and some unrelated staff, so developing the feature was in principle blocked. As I learnt due to some other situations, research
Technical debt and AI: is it gone? Lorenzo Battilocchi May 4 #ai #programming #management #technology 5 reactions Add Comment 3 min read
The pitch for full agentic coding sounds clean: you write specs, agents write code, you review and steer. The human stays "in the loop" as the expert orchestrator. But buried in Anthropic's own research on how AI is transforming work at Anthropic is a sentence that should give every engineer pause: "Effectively using Claude requires supervision, and supervising Claude requires the very coding skil
I’m going on a short vacation this week, so this post is coming out a bit earlier than usual. I actually had a different, more “useful” topic in mind — something educational, something responsible. But then I came across this fascinating article: I don’t like Tailwind. Sorry not sorry written by @freshcaffeine , and I couldn’t get it out of my head. So I decided to write a response instead. I actu
Many companies say people are their greatest asset. Then hire mostly for skills, urgency, and short-term delivery. And later act surprised by friction: • Brilliant people who are hard to collaborate with Often these get treated as individual performance problems. But often they are culture problems. Culture is not the values poster in the hallway. It is the operating system underneath behavior — h
A short build-in-public note from today. We finished shipping the web implementation of Matsuri Platform. Three features now operate as a single integrated flow on the web: Shop (marketplace) Crowdfunding Live Streaming that runs across both — and across in-person events This post is the design retro, not the press release. Watching a stream → wanting the product → buying it → backing the maker's
The problem nobody warned me about Same prompts. Same workflows. Same model. Higher bill — every single week. No input change. No structural change. Just more tokens — and a cost bill I could not explain. What changed under the hood Three things shifted simultaneously, and most cost-increase explanations only mention one of them. 1. The tokenizer itself changed Claude's tokenizer now splits text d