Surveillance is not an AI strategy

AI is a labor bargain before it is a productivity tool.

WIRED reported that Meta is preparing to cut about 10 percent of staff, roughly 8,000 people. That sits beside Meta’s Q1 results release filed with the SEC showing $26.77 billion in net income, $19.84 billion in capital expenditures, and a raised 2026 capital expenditure forecast of $125 billion to $145 billion, partly for additional data-center capacity. WIRED’s account describes morale as historically low, stock-heavy raises getting cut again, and many employees feeling less like partners than costs inside the AI bet.

The human cost is not a footnote. A layoff at that scale is not made gentler because the company is profitable or because the capital plan is ambitious. It tells employees what the company now values, what it will pay for, and which forms of judgment are being recast as expense.

The sharper rupture is the reported laptop activity tracking.

WIRED says Meta’s Model Capability Initiative requires US employees to let the company collect corporate laptop activity, including typing and clicking workflows, to train AI agents. Employees reportedly cannot opt out. The rollout appears more constrained outside the US, where privacy and labor protections are stronger.

If implemented as reported, the policy is not a small implementation detail. It changes the meaning of the tool.

When AI arrives as invisible monitoring, workers do not experience it as help. They experience it as extraction. The company is no longer only asking for output. It is asking for the trace of how work happens, then using that trace to train systems that may later reduce the need for the worker.

Employees understand that bargain. They do not need a philosophy seminar to read the room.

The rest of the reported Meta pattern points the same way. Compensation pressure hits ordinary employees while elite AI researchers reportedly receive packages reaching up to $100 million a year. At least 1,000 top engineers were reportedly moved into Applied AI Engineering, with layoff risk if they refused. Employees called it a draft. Internal AI usage is tracked. Managers feel pressure to drive automation. Workers feel pushed to automate reports, emails, and social product features.

Core AI teams may be energized. That does not erase the other side of the ledger. A lot of people are being asked to live on the cost side of the strategy.

Big-company AI keeps failing the trust test here.

The question is not whether AI can improve work. It can. The question is what kind of relationship the company is building around it. A program that starts with surveillance, automation pressure, and headcount compression will not be read as partnership. Employees will treat it as a machine for making their own work legible enough to remove them.

That reaction is not resistance to technology. It is an accurate reading of incentives.

Automation theater makes the problem worse. A company can track AI usage, pressure managers to raise adoption, and flood internal dashboards with activity while trust collapses underneath. More prompts do not prove better work. More generated emails do not prove better judgment. More agent sessions do not prove a stronger business.

Use visible time returned as the standard, not invisible activity captured.

A good AI system should make the worker’s day less brittle. It should remove retyping, surface exceptions, prepare the packet, catch the missing evidence, draft the routine response, and show where a rule is unclear. The person using it should be able to see the gain. The customer should get a better answer. The business should get a clearer operating standard.

That requires consent and control. People need to know what is being observed, why it is being observed, who can see it, how it changes their job, and what authority they keep. Without those answers, AI becomes a management weapon even when the model itself is impressive.

Hadto’s owner-facing AI has to sit on the opposite side of that line.

No invisible monitoring. No automation theater. No pretending adoption is success because a dashboard went up. No collecting expert practice as raw material while leaving the expert with less power.

The Hadto version starts with named work and visible improvement. A dispatcher should see fewer avoidable callbacks. An estimator should get better job packets. A technician should get clearer standards and less rework. An office lead should see the handoff, the exception, and the next responsible owner. The business should know which rule improved and which promise became easier to keep.

The point is not to turn domain experts into datasets. The point is to turn them into operators and owners.

AI should help capture judgment with the person who holds it, not silently harvest the residue of their work. It should make standards inspectable. It should make apprenticeship easier. It should give people more authority over the workflow they know best. It should return time in ways people can feel and managers can verify without spying.

Small businesses cannot afford to copy the broken version of enterprise AI. They are already thin. They do not have spare trust to burn on covert measurement, fake adoption pressure, or top-down automation campaigns that treat every experienced person as a cost center.

The better SMB playbook is stricter and more human.

Start with the real workflow. Name the promise. Name the owner. Name the evidence. Ask where the expert loses time, where the customer waits, where the handoff fails, and where the rule lives only in someone’s head. Then use AI to make that work teachable, reviewable, and easier to run.

That is a different bargain. The company gets a better system. The worker gets visible time back and more control over the work. The customer gets a more reliable promise. The owner gets a business that depends less on private memory and more on shared operating rules.

Meta’s reported morale problem is not separate from its AI strategy. It is evidence about the strategy.

When AI is introduced as surveillance and compression, employees hear the real message. When AI is introduced as consent-based workflow improvement, visible time returned, and a path from expert labor to operator ownership, people can read that too.

Source material: WIRED, Meta’s New Reality: Record High Profits. Record Low Morale, May 14, 2026, for the employee-reporting, morale, layoff, tracking-program, and AI-adoption details; Meta, Meta Reports First Quarter 2026 Results, April 29, 2026, for net income, capital expenditures, and capex guidance. Reviewed 2026-05-16.