At AI Summit Barcelona today and tomorrow. If you run agents in production, or tried and it did not stick, come find me. Preloop 0.16.0 shipped Monday: flows that start flows under one cost ceiling, notes between agents instead of kill and restart, search any session. #AISB26

On Samothraki, herders carry a hooked stick called gdour. It catches goats by the neck and pole-vaults the herder across cliffs. Catching and leverage tool in one. I'm building that stick for my agents: identity, budget, and a trail of everything they did. Beyond that, they roam.

My conductor agent spent an afternoon arguing with its code reviewer, defending its own slop. Turned out the session had been silently downgraded to Opus. Same goat, same markings, half the brain. First dispatch of Agent Herder, on herding agents the Samothraki way. Link 👇

Do you know what your AI agents did last night? Mine write code, review PRs, call tools and spend real money: Claude Code, Codex, Cursor, Hermes, OpenClaw, OpenCode. Until recently: nothing was watching. So I built the missing layer. 🧵

Most companies are forced to choose between agent autonomy and total control. At Preloop, we’re proving you can have both. This demo shows exactly how we’re making Agentic AI safe for the enterprise. #BuildingInPublic #AIGovernance #MCP

Day 1 of building agents is magic. Day 2 is realizing you're terrified to let them touch production. Today we launched Preloop @ProductHunt. It’s a governance layer for MCP that puts a "breakpoint" on risky agent actions until you approve them from your phone. Demo + Link 👇

Safety shouldn't mean friction. We built Preloop so you can review agent actions from anywhere—your phone or even your Apple Watch. 1. Agent requests high-value action 🛑 2. You tap "Approve" 📲 3. Claude resumes instantly ⚡️ Check out the demo 👇