After the wave of mass layoffs justified by artificial intelligence (AI), several companies are walking it back: rehiring developers. The story going around is that AI didn't deliver the expected result. But the right question isn't why AI failed. The right question is what these companies expected from it.
This post takes apart the easy narrative that the tool let everyone down. What let everyone down was the decision to swap the entire team for the tool. AI keeps delivering exactly what it knows how to deliver. What disappeared was the person who knew how to tell it what needed to be done.
What did companies expect? That AI would understand the business on its own
Let's be direct about the expectation that broke. Companies laid people off en masse and put AI in their place, expecting it to understand the business. The detail nobody factored in: the people who understood the whole business were exactly the ones let go in the layoffs.
It's like firing all the waiters, the cook and the manager of a restaurant, hiring an excellent kitchen robot and expecting it to know how that longtime customer likes their steak cooked, which supplier never runs late and why the house recipe changed in 2019. The robot cooks beautifully. It just has no way of knowing any of that, because that information lived in the heads of the people who left, and nobody wrote it down anywhere.
AI doesn't guess what the business owner can't explain. It doesn't recover context that was never recorded. Expecting that from it isn't using AI, it's betting that the tool will fill a hole the company itself dug.
The fault isn't AI's, it's a sequence of bad choices
Blaming AI is comfortable because it takes responsibility off whoever decided. But the sequence of choices is clear when you look at it from the outside.
First, the company got rid of the team that knew everything: the history of the decisions, the exceptions, the reasons why, the difficult clients, the parts of the system that can't be touched without breaking something else. Second, in most cases that team left without documenting anything, because documentation was rarely a priority while people were still around to remember it all. Third, it handed AI the mission of producing results out of something not even the owner could define precisely.
None of those three choices is the technology's fault. They are management decisions. And their sum produces the obvious result: with no one who knows the business and no record of what was built, AI generates code that works today and that no one understands tomorrow. We already gave this effect a name on our blog: it's cognitive debt, the hidden cost of speeding up the writing and forgetting to keep the understanding alive.
AI becomes an expert on your product, but that takes time and coverage
Here's the part almost no one took advantage of. AI is already an expert in code. And it can become an expert on your product too, on your way of operating, on your rules, on your exceptions. It just doesn't come that way out of the box. It comes from training.
That training has two ingredients that only the people on the inside can provide: time and coverage. Time is the accumulation of context, every decision explained, every exception recorded, every reason written down. Coverage is how far that reaches across the whole operation, every developer documenting the part they know best, turning knowledge that lived only in someone's head into something AI can use. With time and coverage, AI stops being a generic tool and becomes the living memory of the product, the one that never takes a vacation and never quits.
The companies that laid people off en masse did the exact opposite. They cut precisely the people who would provide that time and that coverage. They killed the source of the training and then complained that the student never learned.
AI executes the goal. The one who sets the goal is human
There's a confusion of roles at the bottom of this story, and it needs to be clear. AI delivers results when the goal is well defined. Give it a clear objective, with context and limits, and it executes at a speed no human team reaches on its own. That's its superpower.
What it doesn't do, and shouldn't do, is make the decision. Deciding what the business needs, what the priority is, which risk is worth taking, what truly serves the customer, that's human work. AI is an extraordinary executor of goals and a terrible owner of decisions, because a decision demands responsibility, business context and judgment, and none of that is a tool's job.
When a company fires the people who set the goals and keeps only the ones who execute, it flips the logic. It's left with the executing machine and no driver. AI starts receiving poorly defined goals, or no goal at all, and hands back exactly what it received: work with no direction. Then comes the rehiring, because someone has to go back to saying where to go.
The path they missed: empower the team, don't replace the team
There was a much better choice on the table, and it didn't cost the entire team. It was using AI to empower the developers who were already there. The same ones who knew the business, now delivering in weeks what used to take years, because AI takes the grunt work off their hands and gives back time for what only humans do: think, decide and explain.
In that scenario, each person on the team becomes a manager of context. While delivering faster, they feed AI the time and the coverage that turn the tool into the true expert on the product. The team doesn't shrink, it gets stronger. And the knowledge doesn't leave with whoever leaves, it stays recorded and available.
This is exactly the work we do at Steply. We don't swap people for AI. We put AI to work multiplying the right people and we capture their knowledge in a way the tool can use, so the company accelerates without digging its own cognitive debt. Faster delivery today, without losing the understanding tomorrow.
The reframe that separates who wins from who rehires
The wave of rehiring isn't proof that AI failed. It's the bill for a wrong decision coming due. Whoever swapped the entire team for the tool is paying twice: they paid for the layoff and now they pay for the return, with the knowledge lost somewhere along the way.
Whoever understood the game did it differently. They kept the people who know the business, gave them AI to go further and faster, and used that time to train the tool on what's specific about the product. AI isn't the new owner of the decision. It's the best executor your team has ever had. The difference between winning and rehiring is in who you left in command.