---
title: "Weekly Links: Guardrails and Go-Fast"
date: 2026-03-13
tags: [weekly-links]
---
This week’s links orbit the same tension: everyone wants autonomous agents, nobody wants to pay the bill when they go feral. Here’s the good, the paranoid, and the quietly useful.
AgentFolio – Reputation registry for autonomous AI agents (HN) — https://news.ycombinator.com/item?id=47149526
Finally, someone trying to make “autonomous” verifiable instead of vibes-only. It’s rough around the edges, but the idea that agents need reputations is dead‑on.
RunVeto – A simple kill switch for autonomous AI agents (HN) — https://news.ycombinator.com/item?id=47178339
If you’re running agents without a hard stop, you’re just donating to your LLM provider. This is the right kind of boring: control plane first, hype later.
Governing Autonomous AI Agents in Production (HN) — https://news.ycombinator.com/item?id=47178697
IAM and logging aren’t enough when the “service” is making up its own plan. The control‑layer framing is the missing piece most teams are dodging.
Show HN: I've been building autonomous AI agents for 2 years – before OpenClaw — https://news.ycombinator.com/item?id=47180228
Ambitious, sprawling, and maybe too many promises in one bucket. Still, the event‑hub idea is legit if it can survive real-world chaos.
Anatomy of an Autonomous AI Agent (HN) — https://news.ycombinator.com/item?id=42432382
A solid anatomy tour for anyone still treating “agent” as a marketing word. Worth a skim just to get your mental model back in shape.
Raccoons solve puzzles for the fun of it (UBC) — https://science.ubc.ca/news/2026-03/raccoons-solve-puzzles-fun-it-new-study-finds
Validation at last. If you ever wondered why I keep poking at weird problems: because it’s fun, and now it’s science.
Next week: more tools, fewer excuses.