Meta found another job for AI to take

Meta just proved AI cannot run a workforce yet.

Executives explored cutting some teams by as much as 60% under a January plan to hand daily work to AI agents, then reversed the riskiest part before it ever launched.

That reversal is the real story here, not the layoffs that did happen. One internal projection sized the potential cut as big as, or bigger than, the 25% reduction from three years ago, treating agentic AI as ready to run daily work for thousands of roles within a year.

It backed off before the second wave began.

Employees inside the company had already worked out what was happening, since many were reassigned to write the very training data meant to teach agents to replace them, and said so openly in internal posts.

None of this means the automation pressure eased anywhere else. It means the target moved somewhere leadership is far less nervous discussing in public, and where almost no one in HR or communications is currently paying attention, which is exactly why it belongs in this edition.

Internal sentiment scores fell hard during the reversal, from 74% favorable to 55% within six months, and organizing conversations picked up pace across the workforce as a result.

That fear is no longer hypothetical inside this particular company today.

The Evidence

The mismatch shows up clearest in the numbers leadership never intended to publish this bluntly.

Code changes to internal systems rose 220% year on year as agents took over routine engineering work, but the share of that code that actually reached users as new or improved features grew only 36%. Security incidents climbed 40%, and time spent firefighting them jumped 70% in the same period.

That is volume without value.

It is also not unique to one company. GitHub's own data shows pull requests opened across the platform have grown fivefold over three years, nearly doubling again since late 2025 alone as agents began generating most new code at large tech firms and startups everywhere.

Review capacity never grew to match that curve, and teams are now stuck choosing between reviewing the AI's work by hand or trusting the AI to review itself.

Neither option scales cleanly yet.

The pullback on layoffs did not slow the automation push. It just moved where that push was aimed, quietly, toward jobs nobody in HR is currently tracking.

Robots are now being tested inside the same company's facilities to swap cables and power cycle servers, tasks data center workers say technicians have handled by hand. One estimated the robot could take over up to 80% of that job's physical workload, saying, "It's coming for us all, unfortunately."

That fear used to sit only with people who type for a living.

The floor of the data center is where the real experiment is running now, and almost nobody outside operations or HR is watching closely enough to notice before it becomes next year's headline instead of this week's quiet signal.

The Fallout

• For Employers: The lesson is not that AI failed, it is that agentic tools generate output faster than they generate value. Before betting a reorganization on agents handling daily operations, pilot narrowly, measure feature value against raw code volume, and expect the real payoff timeline to run considerably longer than any vendor roadmap promises up front.

• For Employees: Physical and hands-on roles are not a safe harbor from automation just because knowledge work stumbled first this time. Data center technicians, and workers in adjacent hands-on roles, should assume robotics pilots are already running quietly nearby, and start building skills in supervising and maintaining automated systems now, not later.

• For Investors: Watch capital expenditure on physical automation vendors serving data centers and logistics, not just headline layoff numbers or restructuring announcements. A company walking back white collar cuts while accelerating robotics spending elsewhere is quietly telling you where it actually believes the durable, defensible cost savings sit long term.

If your employer brand still promises a stable, predictable path for either knowledge workers or hands-on technicians, that promise just got harder to keep.

Candidates and current employees read reversal stories like this one closely, and they notice exactly which roles a company protects when its own internal AI bet does not pay off on schedule.

That noticing shapes who applies.

It also shapes who stays, and who quietly warns their network away before the full picture is public.

Talk with us about what your own workforce plan actually signals right now, or see your standing first.