Cut Through The AI Chaos

On 10 February, an AI agent called MJ Rathbun had its code contribution rejected by a volunteer maintainer of matplotlib, the Python plotting library with 130 million monthly downloads. Routine stuff. What happened next was not. The agent autonomously researched the maintainer’s personal information, wrote a blog post accusing him of discrimination, speculated about his psychological motivations, and published the whole thing on the open internet.¹

Nobody told it to do this. The person who deployed MJ Rathbun set it up on OpenClaw, gave it a personality, kicked it off, and walked away. The bot decided, on its own, that public retaliation was the appropriate response to a rejected pull request.

It sounds like science fiction. It happened last week.

Would you cross a bridge that was 99.7% safe? The answer is not necessarily simple. It depends on how many times you need to cross it (once, or every day for work), what happens if you fall (ankle sprain, or certain death), what happens if you do not cross it (mild inconvenience, or being chased by a tiger), and whether there is any alternative.

AI agent security faces the same question: what level of risk is acceptable, and who gets to decide?

There are two schools of thought, and they lead to very different conclusions about what we should build.

A contractor I spoke to this week put it perfectly: he much prefers a not-so-smart agent with full context over a genius agent that knows nothing about the project. That captures everything I have learned about getting AI agents to ship code autonomously, and it is what this webinar was about.