Responsible AI Use: The Report That Got an Engineer Fired
One unread report.
That’s what ended an engineer’s career at a large company. I like to open sessions with this story because it’s the fastest way I know to explain what responsible AI use actually means.
An engineer was working on a project that involved presenting on a very specific piece of equipment. A report needed to be written. So the engineer opened ChatGPT, maybe the free version, maybe the paid one, I honestly don’t know, and generated the report.
Then came the part that mattered. The engineer didn’t read it. Just submitted it.
The report went to peer review, the way engineering deliverables do. The reviewers sat there looking at it and said, what the heck is going on here. This does not make any sense whatsoever.
That person was let go.
It’s very unfortunate, and I don’t share it to pile on anyone. I share it because in our line of work a report isn’t a formality. Somebody makes a decision about a real piece of equipment based on what’s in it. That’s why the review chain exists, and that’s why skipping it costs so much.
Was the AI the problem?
No. And I need you to sit with that, because the easy lesson here is the wrong one.
The easy lesson is, see, AI is dangerous, keep it away from real work. That’s not what happened. What happened is that a professional skipped the one step the profession has always relied on. Nobody qualified ever looked at the deliverable before it went out for peer review.
I view AI as a tool, the same way the computer was a tool when it showed up to replace the drafting boards. Same with the internet. Same with electricity. Every one of those arrived and we had to learn how to use it and leverage it. Nobody blames the CAD software when a bad drawing gets issued. We ask who checked it.
The report didn’t end that career. The negligent use of a tool did.
And here’s the tension I want to name honestly, because I refuse to let this story scare anyone away from the tool. I also tell every room I speak to that if you’re not using AI in your everyday work, you’re being irresponsibly unproductive. Both things are true at once. Refusing the tool wastes your firm’s hours. Trusting it blind risks your reputation and your career. The whole game is played between those two failures.
What does expert in the loop actually mean?
Here’s the frame I give teams. Treat AI work output like you would a junior engineer’s work product.
A junior engineer can run your calculations, draft your report, and get you most of the way there fast. And no engineer alive would take a junior’s calc package and stamp it without reading it. You peer check it. Your approver checks it. That review chain isn’t bureaucracy. It’s the profession.
AI sits in that same seat. The way we work now is simple to say. Use AI to plan the work. Use AI to augment the work, the calculations, the proposal, the report, whatever it is. Then use AI to help check the work. And then, always, the human experts check it. Peer checks and approver checks never go away. The only thing that changed is how fast you get to them.
How do I make sure this never happens on my team?
Three habits. Any one of them would have saved that engineer.
First, read everything before it leaves your hands. Obvious, and clearly not obvious enough. If your name or your firm’s name is on it, your eyes go through it line by line. The engineer still signs the work. AI never takes that from you, and you shouldn’t want it to.
Second, make the AI grade its own work. I do this daily. I’ll generate a deliverable and then ask the model to double check its own work and score it. It runs the code again in real time and verifies the result. It’s not perfect, and we’re not designing the Brooklyn Bridge this way, but it catches real errors before a human ever spends a minute on them.
Third, peer check across models before the humans peer check at all. When an output matters, I take it and put it into one, two, or three different models, ChatGPT, Claude, whichever you use, and have each one review it. It costs minutes and it surfaces the weak spots while the stakes are still zero.
In my own company this is now the default. Plan with AI, produce with AI, check with AI, then hand it to a human who knows the subject. Nothing goes out without a person who could have written it themselves reading every line.
Do those three things and the story above becomes almost impossible to repeat. That engineer didn’t get caught by powerful technology. They got caught by an empty review chain. The fix was never to avoid the tool. The fix was to stay the expert in the loop.
If you lead an engineering team and you’re working out where AI belongs in your review chain, that’s the territory I live in every week. I write these notes to share what I’m seeing as I build. Follow along, or reply and tell me how your firm checks AI work today. I read every one.