Drinks with an Engineer Who Said "AI Is Useless"

Earlier this year I caught up with an engineer friend over drinks. We hadn’t seen each other in a while, and somewhere between the second and third round, AI came up.

He didn’t hesitate. “I don’t use AI. It’s useless, right?”

When I asked why, his answer was short. He’d tried ChatGPT for a bit, and it didn’t work for him. End of story.

Honestly, I understand that reaction. I’ve been there myself. But the longer I sat with what he said, the more I realized it was the perfect snapshot of a divide I keep running into.

”I tried it a bit” almost always means you didn’t really try

In my experience, attitudes about AI fall cleanly into two camps, and the dividing line is how much time someone has actually put in.

• People who tried it for an afternoon land on “it’s useless.”
• People who used it seriously for weeks land on “this is insane.”

There isn’t much middle ground. I started on the skeptical side too — my first impression was lukewarm at best. What changed wasn’t the model. It was how I used it.

The skepticism is reasonable

I don’t think this is just about being behind. Engineers have lived through a parade of overhyped tools that promised the moon and delivered a footnote.

• Automation tools that didn’t automate the parts that mattered
• Low-code platforms that broke down the moment requirements got real
• Frameworks that owned the conference circuit one year and disappeared the next

So when something new shows up wearing the same costume, “isn’t this just another one of those?” is a perfectly rational question.

And to be honest, today’s AI still isn’t perfect. The way I’d put it:

It’s less like AI writes code for you, and more like you’re babysitting the AI while it writes code.

Leave it unattended and it drifts. Real use looks like a constant loop of giving direction, correcting drift, and steering it back on course. That’s the job now.

But this time really is different from the old pipe dreams

Here’s what makes this wave different, and it’s the part skeptics tend to miss:

It actually works. Not for everything, but for enough.

It doesn’t deliver a 100. It reliably delivers a 70. And in the messy reality of day-to-day work, “reliably 70” is enormous.

A few places where I now reach for AI by default:

• Generating test code from a function signature
• A first-pass SQL query I can refine instead of write from scratch
• The opening 30 minutes of any research task
• Drafting documentation I’d otherwise put off

For work like this, it’s no longer a question of whether AI is useful. It’s a question of why anyone would do it the slow way.

There’s another reason some engineers don’t change

The friend I was drinking with had spent years on the same project, with the same team, in the same stack. He didn’t have much exposure to what was happening outside that bubble.

When you’re inside that kind of environment:

• You don’t bump into new tools naturally
• Nobody around you is using them either
• There’s no felt urgency to change

And honestly, that’s fine — if the world around you stays the same. If you’re going to spend the rest of your career in that exact context, you can keep doing what works.

The risk is that environments rarely stay the same. One day your client decides to roll out AI tooling. One day a younger engineer ships in two days what used to take you a week. One day the comparison stops being theoretical.

That’s how people end up on the “we don’t really need you anymore” side of the table. Not through any single failure, just through standing still while the floor moved.

I didn’t try to talk him into it

I didn’t push. We’re not close enough for that, and unsolicited career advice over drinks rarely lands well anyway.

I just said one thing before changing the subject: “AI actually works pretty well, if you use it properly.”

That’s all I had to offer. What he does with it is up to him.

This isn’t about right or wrong. It’s about timing.

I keep coming back to this framing. It’s not that the people who dismiss AI are wrong, exactly. They’re just early or late.

• Notice early, and it becomes a real edge.
• Notice late, and it becomes the gap between you and the people who did.

The interesting question isn’t whether AI is useful. It’s when you personally decide to find out.

The honest, working stance

AI isn’t magic, and treating it like magic is how you end up disappointed. But treating it like nothing is how you end up behind. The version that works in practice is somewhere in between:

• ❌ Hand the work off to AI and hope
• ⭕ Stay in the driver’s seat and use AI to go faster

Get that framing right and the productivity gain is genuinely large. Get it wrong in either direction and you’ll either ship garbage or refuse to ship at all.

Summary

• “AI is useless” is an understandable first reaction
• But sustained, serious use almost always flips the verdict
• It isn’t perfect, but it’s a real force multiplier
• Some environments make it hard to even notice the shift
• There’s no point pushing people who aren’t ready

I expect this conversation to keep happening for a while. And quietly, in the background, the people who decided to find out are pulling ahead of the people who decided not to.