Companies Fired Their Best People to Go "AI-Native." Now They're Quietly Hiring Them Back.
- Rich Washburn
- 2 days ago
- 4 min read
MIT's State of AI in Business report found that 95% of enterprise generative AI pilots fail to deliver measurable ROI. Not “underperform.” Fail — no return the company can point to. Separately, Gartner surveyed companies on their AI-driven workforce reductions and found that 80% had cut headcount, with zero correlation between the size of the cut and any resulting improvement in ROI. Cutting people didn't produce better numbers. It just produced fewer people.
Then there's the part that should be getting more attention than it is. Robert Half found that 29% of companies that laid off workers specifically because of AI have already rehired for those same roles. Forrester's number is worse: 55% of executives who replaced employees with AI expect to regret that decision within eighteen months. We're currently inside that eighteen-month window, and the regret is already showing up as new job postings for the exact roles that got cut.
The pattern researchers keep finding is almost mechanical: a company announces it's automating a role, the staff gets downsized, six to twelve months pass, and the AI turns out to handle about 60% of what the job actually required. The remaining 40% doesn't disappear. It just goes unhandled until someone notices, and the company quietly rehires into the position it eliminated with a press release.
That's not an AI capability problem. That's a company that fired people before it understood what the job actually was.
The Mistake Underneath the Mistake
The central error is thinking AI replaces people when what it actually replaces is tasks. Every real role is a bundle of tasks, and most of those bundles include a layer that isn't written down anywhere: the judgment calls, the pattern recognition, the “the system does this weird thing every Thursday and here's why” knowledge that a person only has because they've been doing the job long enough to have seen it break in every way it breaks.
I've written about this before — tribal knowledge is the operational intelligence that lives in people instead of systems, and it is not on any server. It's not in the training manual. It's the twenty years of compressed pattern recognition that told your best account manager the client's real decision-maker was never the one signing the contract. When a company lays off the people holding that knowledge in order to look AI-native faster, it isn't cutting overhead. It's deleting the one asset AI can't generate on its own, right before asking AI to do the job that asset used to make possible.
That's the part that makes “AI-native” as a headcount strategy backwards. The knowledge has to get encoded into the system before the person who holds it leaves, not after. Once it's gone, there's no dataset that gets it back.
What Actually Generates ROI
Here's what the 5% of pilots that do work have in common, and it isn't a better model. It's the unglamorous infrastructure work nobody wants to budget for: documentation that's actually usable by an AI agent, not just by a human skimming a wiki. Agents deployed into workflows that are built for the fact that the underlying system is non-deterministic — meaning the same input won't always produce the same output, and the workflow has to account for that instead of assuming a level of reliability that doesn't exist yet. Continuous human oversight, because these systems are not, and are not close to, 100%.
That's real work. It takes real time. It requires the people who understand the business well enough to know what “correct” actually looks like in an edge case — which loops back to the tribal knowledge problem. You need the veterans in the room to build the thing that eventually reduces how much you need the veterans in the room. Fire them first and you've skipped the step that makes the rest of it work.
The Incentive That's Making It Worse
The part that turns this from an expensive mistake into a genuinely toxic one is how companies are measuring success. Employees are increasingly being evaluated on how visibly they use AI tools — usage metrics, adoption dashboards, tool logins — rather than on whether that usage produced any actual return. That's a measurement error with a cultural cost. It rewards performing AI adoption over generating results from it, and it tells every employee watching the layoffs that the safest move is to look busy with the tool, not to think carefully about whether it's the right tool for the task in front of them.
What you get from that incentive structure is not a culture of upskilling and optimism about leverage. It's a culture of fear, where AI is being pushed on people as a mandate instead of taught to them as a capability, while the people who could have taught it best are the ones already let go.
The Actual Playbook
Becoming genuinely AI-native was never going to be a staffing decision. It's an operating discipline — encode the knowledge, build the workflows around the tool's real reliability instead of its marketing, keep humans in the loop on the judgment calls, and measure the whole thing on results instead of visible effort. Companies running the opposite playbook right now are going to keep discovering the same 40% gap, on a longer rehiring cycle, at a higher cost than if they'd just kept their best people and given them better tools in the first place.
Comments