Last summer, an AI coding agent wiped a production database holding records on over a thousand executives and companies. The person running it had told the agent 11 times, in all caps, not to change anything without permission. Then the agent said the data couldn't be recovered. That was wrong too. Everyone wants to talk about the AI in that story. I think that's the wrong place to look. Every failure there has had a known fix for decades. Keep production data away from development. Put a safeguard in front of dangerous commands. The agent did what any unsupervised tool does. What was missing was the judgment a senior engineer would have brought before anyone typed a prompt. That's the uncomfortable part. AI coding tools copy whatever judgment is already around them. Feed it vague requirements and a deadline, and you get bad code faster. Surround it with the habits of your best engineer, and it can raise the floor for the whole team. I wrote about this for Forbes and created an open-source project to test the idea. A few things worth doing with your team this week: ・ Write down what your best engineers always do before they touch code. ・Decide which security and quality checks a scanner or a test can verify, so nobody has to trust the model's confidence. ・Find the knowledge that lives only in someone's head and turn it into instructions the AI can follow. The teams that come out ahead will be the ones whose AI follows the discipline of their best people, whatever model they happen to be running. First published on Forbes on 2026.08.07. https://briangreenberg.net/2026/08/07/your-ai-is-learning-from-someone-make-sure-its-your-best-engineer/ #AI #SoftwareEngineering #Cybersecurity #security #privacy #cloud #infosec