The AI Productivity Lie

You never had an output problem

Most orgs don’t have a shortage of output. They have a shortage of good ideas.

The fact that implementation was expensive was quietly doing you a favor by filtering out the bad ones. When something costs six weeks and four engineers, you think about it. You argue about it. You kill it if it’s not worth it.

Now the cost of producing something is basically zero. So that filter is gone. And what’s coming through isn’t innovation. It’s half-baked features nobody asked for, internal tools nobody needed, and dashboards that will be opened exactly once.

Thomas shipped eleven internal tools in Q4. Eleven. I asked how many were still in use. He checked. Two.

Nobody wants to be a 10x engineer

There’s something nobody wants to say out loud.

The majority of your workforce isn’t trying to become 10x engineers. They want to do their job, clock out, and go home. That’s not a moral failing. That’s just reality.

They’re not using AI to be dramatically more productive. They’re using it to do the same work with less effort. The output hasn’t changed. The energy has.

I told Thomas this and watched him try to argue. He couldn’t. He’d seen the Jira boards. Same velocity. Same throughput. People just seemed slightly less tired in standups.

Which might actually be the best outcome. But it’s not the one he put in his board deck.

Your best people are going to leave

This is the part that actually hurts.

The two or three people on Thomas’s team who actually cared about craft? The ones who took pride in clean architecture, code that would still make sense in six months? They were miserable.

They were drowning in AI-generated slop they were expected to review, fix, and maintain. Pull requests coming in faster than ever. Quality through the floor. Every review turned into an archaeology dig. Is this a real design choice or did someone just hit tab on whatever Copilot suggested?

These are the people holding your codebase together. They’re going to burn out and leave. Not because of AI. Because you turned them into janitors for a machine that doesn’t know what good looks like.

You’re still stuck in traffic

Even when work genuinely gets produced faster, you’re still behind the same walls you’ve always had.

Approvals. Dependencies. Compliance. Stakeholder alignment. The designer is on vacation. The PM changed the requirements. The API team hasn’t finished their migration.

Making the coding part faster is giving someone a faster car and then watching them sit in the same traffic.

Thomas laughed at this. Then stopped laughing. His compliance team had a three-week backlog and his fastest squad had been blocked for nine days on a security review.

The bill

One more thing.

Thomas’s CFO did the math. Each engineer now costs an extra €2,000/month in LLM subscriptions. Forty engineers. That’s €80,000 a month. Close to a million a year.

For what? Same output. Less effort. Burned out seniors. Same bottlenecks. Nine out of eleven tools sitting there unused.

He stared at his coffee for a while after that one.

So what do you do

Fix your approval process. Unblock your dependencies. Give your senior people room to breathe. Build a culture where killing a bad idea early is celebrated instead of punished.

Then think about where AI actually helps. It does help. In specific, targeted, well-understood ways. Just not in the magical org-wide “everyone becomes 10x” way you were promised.

Thomas closed his laptop and said “So basically I spent a fortune automating the part that wasn’t broken.”

Yeah. Pretty much.