Boomerang Hiring: Two Thirds of AI Layoffs Are Being Reversed

Boomerang hiring, the rehire of someone a company previously let go, is climbing faster in tech than in any other sector. Two thirds of the companies that cut staff for AI reasons have already rehired a quarter or more of the roles they eliminated.
 

The cuts are why those AI rollouts underperformed, which changes what you check before the next restructuring decision.
 

The Cut That Cost More Than It Saved

Roughly 55,000 job cuts in 2025 were attributed directly to AI. The reversals started arriving inside the year.

THE REVERSAL IN THREE NUMBERS

Spending more to rebuild than you saved means the savings were never there.

Some of those reversals bring the person back. Forrester expects a share of them to reappear offshore or at lower wages instead, which recovers the headcount without recovering what left with it.
 

The Knowledge Nobody Wrote Down

The written procedure covers how to process the claim. It does not cover which claims come back as fraud, which customer needs a manager on the first call, or which vendor disputes every invoice and needs documentation attached upfront.

Underneath every documented process sits a second version: the conditions where the policy gets overridden, the cases a twelve-year veteran catches on sight, the things the company will never do that nobody wrote down. A system trained on the documented version handles the documented version. Production is mostly the other one.

Klarna is the clearest public case. Roughly 700 customer service roles cut on the projection that AI would absorb the volume. The volume held. Disputes and fraud cases, anything needing a judgment call, overwhelmed the system. Rehiring cost more than the original savings projection. 

One in Five of Google’s AI Hires Had Worked There Before

Twenty percent of the software engineers Google hired into AI roles in 2025 were former employees returning.

Google has a deep alumni pool after its 2023 layoffs and is competing with OpenAI, Anthropic, and Meta for the same engineers. A returning engineer already knows the internal tools, the codebase, and the approval chain, which is months of context a new hire has to build. That context is the reason the alumni list is worth going back to, and it never appeared in anyone’s efficiency case.
 

The Rehire Puts It Back in One Head

The exception logic returns to exactly where it was, in one or two heads, written down nowhere. The company has repurchased its original single point of failure at a higher price, and the next resignation deletes it again on the same terms.
 

Three Things to Check Before the Next Cut

1. Open the SOP for your three highest-volume processes and mark every step that says “escalate” or “use judgment.” Those steps are where the undocumented knowledge lives. Count them.
2. Ask whoever owns those steps for the last five cases that went off-script. You want the five actual cases rather than a summary of them. They are the design requirements the written procedure does not contain.
3. Count what share of last month’s volume needed a call outside the documented path. Compare it to the share your pilot ran on. That gap is your exposure, and it is the number the efficiency case should have been built on.

Where the Knowledge Should Actually Live

All three checks work until the quarter it matters, when the restructuring is already moving and attention is scarcest.

The durable version holds that knowledge in a system rather than a person. Override conditions, exception rules, the constraints nobody wrote down, kept as a live layer the company owns regardless of who stays. That is what the Grip does inside Mustang: the decisions senior people make from experience, documented for the first time.