I build with AI every day, and I have pushed my teams to do the same. We have rolled out AI tooling across engineering, built production agents for the boring work, and I spend real time coaching engineers on how to get more out of these tools. I am not a skeptic. I think this is the biggest shift in how we build software in my career.

So I want to be careful about how I say the next part, because it is not a complaint about AI. It is a warning about what AI does to us.

It is easier than ever to trick yourself into thinking you can do something. I felt it myself during a hackathon. I built an application over a weekend that genuinely impressed people, and impressed me. It felt like magic. But there is a wide gap between a demo that impresses a room and a system you would actually stand behind, and AI is very good at getting you across the first gap while hiding the second. The dopamine arrives long before the real work is done.

I have seen the same thing show up in ordinary work products. The project plan that is ten times longer than it needs to be. Confident, well formatted, but oddly twisted, redundant and filled with facts that have quietly drifted away from true. It reads well. That is exactly the problem. Slop is plausible, polished. It passes the skim test. Polished and wrong is dangerous. Polished and wrong gets waved through when what we needed was accuracy.

Some of what’s changed is where the cost sits. Producing volume used to be expensive, so a thick document was a decent signal that someone had done the work. Now volume is free. The scarce thing is judgment: knowing what good looks like, and being willing to throw away three pages of confident text because it is subtly off.

When I coach an engineer who is getting bad output, the problem is almost never the model. It is that they did not understand the problem well enough to ask for the right thing. The work did not disappear. It moved. You still have to build a real grasp of what you are doing, or the machine will happily hand you something that only looks like what you need.

So yes, I want my teams using AI aggressively. I also want them to be the people who can tell the difference between something that works and something that just looks like it. That discernment is not a nice-to-have. It is the job.