Meta’s attempt to rebuild its workforce around artificial intelligence has exposed a problem that extends far beyond one company: replacing human work with AI is considerably harder than reducing the number of people doing it. What began as an ambitious effort to create an AI-native organization evolved into layoffs, employee resistance, operational problems and a retreat from the most aggressive parts of the plan. The episode suggests that the immediate obstacle to AI-driven corporate restructuring is not simply whether the technology can perform individual tasks, but whether it can reliably reproduce the broader systems of judgment, coordination and accountability that make organizations function.

The initiative, known internally as Project OT, was designed around a radical proposition. Instead of merely giving existing employees AI tools, Meta explored restructuring teams so that autonomous AI systems could perform much of the routine work while smaller groups of highly skilled employees supervised them. Internal scenario planning considered reductions of up to 60% in some teams, although Meta has stressed that this did not mean a 60% reduction in the company’s overall workforce. The plan eventually produced a roughly 10% workforce reduction, while a second restructuring wave planned for later in the year was abandoned.

The distinction is important because it reveals what went wrong. Meta was not simply experimenting with productivity software. It was attempting to redesign the organization around the assumption that AI agents would become sufficiently capable to replace substantial amounts of human coordination and execution. That assumption proved premature.

Automation Advanced Faster Than Reliability

Meta’s AI-native strategy emerged from a broader belief in Silicon Valley that generative AI could fundamentally change how companies are organized. Traditional corporate structures divide responsibilities among engineers, designers, product managers, analysts and managers. Meta’s proposed alternative was built around smaller, more flexible teams in which employees would become general-purpose builders supported by AI systems and autonomous agents.

The logic was straightforward. If AI could generate code, analyze information, produce prototypes and eventually execute increasingly complex tasks, fewer specialists would theoretically be needed. Smaller teams could move faster, middle-management layers could disappear and highly productive employees could oversee AI systems capable of doing the work of much larger groups.

The difficulty was that output is not the same as productivity. Internal figures cited in the reporting showed that AI-assisted coding generated a dramatic increase in code changes inside Meta, but the increase in changes reaching users was far smaller. Code changes to internal software platforms and infrastructure reportedly rose 220% year over year, while changes producing new or upgraded features for users increased 36%. The gap illustrates a fundamental weakness in measuring AI productivity: generating more material is easy to quantify, while determining whether that material creates useful economic value is much harder.

That distinction becomes even more significant when AI is given greater autonomy. Internal reports described reliability problems and a rise in disruptive actions by AI agents. Technical and security incidents reportedly increased, while employees spent substantially more time dealing with the consequences. In such an environment, automation does not eliminate human work. It can instead redistribute that work toward monitoring, correcting and repairing machines that were supposed to make organizations more efficient.

Layoffs Turned Employees Into AI Opponents

The second weakness was organizational rather than technological. Employees were asked to embrace systems that appeared increasingly connected to the possibility of eliminating their own jobs. That created an obvious incentive problem.

The message was particularly difficult to sell because layoffs, restructuring and AI adoption were happening simultaneously. Engineers were moved into new AI-related roles, some of which involved producing training data for Meta’s models, while other employees were being dismissed. At the same time, the company introduced systems intended to capture how employees interacted with computers so that AI could learn to reproduce those activities. To workers already worried about job security, the distinction between “AI augmentation” and “AI replacement” became increasingly difficult to believe.

The result was a collapse in internal confidence. Meta’s employee sentiment measure reportedly fell from 74% favorable to 55%, while employees publicly challenged executives and mocked elements of the transformation. The reaction was not merely emotional resistance to technological change. It reflected a rational concern that workers were being asked to help build systems whose long-term purpose could be to make their positions unnecessary.

This is where Meta’s strategy encountered a contradiction. AI requires enormous amounts of human participation before it can become dependable. Models need data, evaluation, supervision, testing and correction. The very employees whose jobs are being redesigned or eliminated often possess the institutional knowledge required to determine whether an AI-generated result is actually useful.

Removing those people too quickly can therefore weaken the system being automated. A company can reduce headcount and increase AI usage at the same time while becoming less productive if the remaining employees spend more time supervising flawed outputs and resolving failures.

Meta’s eventual decision to cancel the second restructuring wave showed that management recognized at least some of these risks. The company proceeded with the initial 10% reduction but subsequently emphasized greater stability for remaining employees. Zuckerberg later acknowledged that AI agent technology had not progressed as rapidly as he had expected.

The Financial Pressure Made the Experiment Harder

Meta’s workforce restructuring also cannot be separated from the extraordinary financial commitment behind its AI ambitions. The company is investing heavily not only in models and software but also in data centers, chips and computing infrastructure. In July, Meta raised its 2026 capital expenditure forecast to between $130 billion and $145 billion. Its second-quarter free cash flow fell 91% year over year to $784 million, highlighting how aggressively the AI buildout was consuming cash.

That creates pressure for AI to demonstrate returns quickly. The more a company spends on infrastructure, the stronger the incentive becomes to show that AI is improving productivity, creating new revenue or reducing costs. Workforce reductions can appear attractive because they offer an immediate and visible financial benefit.

But layoffs also create a dangerous temptation: treating headcount reduction as evidence that automation has succeeded. The two are not equivalent. A company can reduce employees before the technology is capable of replacing their contribution. In that case, the organization may simply become smaller without becoming proportionately more productive.

Meta’s broader AI spending makes this distinction particularly important. The company continues to argue that AI agents could eventually become a major business, while analysts and investors have increasingly focused on whether the enormous infrastructure investment can generate sufficient returns. Reuters reported in July that Meta’s free cash flow had fallen sharply as spending expanded, while the company continued to defend AI as a long-term opportunity.

The pressure is therefore coming from both directions. Employees want evidence that AI will not simply become a mechanism for eliminating jobs, while investors want evidence that the billions being spent on AI will eventually produce substantial economic returns. Management has to satisfy both groups before the technology has fully matured.

Meta’s Retreat Points to a Different AI Workplace

The most revealing aspect of Project OT may therefore be that Meta did not abandon AI. It abandoned the idea that organizational transformation could move at the same speed as AI enthusiasm.

Meta continues to invest heavily in AI infrastructure, models and agents. It is expanding computing capacity, developing its own AI chips and pursuing new AI products. The company is clearly not retreating from the technology itself. Instead, the retreat has been from the most aggressive assumption about what AI can immediately do inside a large organization.

That distinction could define the next phase of corporate AI adoption. The successful model may not be one in which companies suddenly replace large numbers of employees with autonomous systems. It may be one in which AI gradually absorbs specific tasks while humans remain responsible for judgment, quality control, security, relationships and accountability.

For Meta, the lesson is especially significant because the company attempted to move directly toward an AI-native organization while the technology was still developing. The experiment revealed that organizational complexity cannot be automated simply by placing an AI agent inside a smaller team. Human workers do more than complete discrete tasks. They interpret ambiguous information, understand institutional history, identify unusual risks and decide when a machine’s answer should not be trusted.

That does not mean AI will fail to reduce employment or transform corporate structures. It means the timing and sequence matter. The technology has to become reliable enough before organizations can safely remove the human systems that currently compensate for its weaknesses.

Meta’s experience therefore offers a more complicated picture of the AI revolution than the simple narrative of machines replacing workers. The immediate challenge is not proving that AI can perform human tasks. It is proving that AI can perform enough of them, reliably enough, while preserving the coordination and accountability that businesses depend on.

Project OT did not demonstrate that AI cannot transform Meta. It demonstrated that transforming an organization around AI is itself a much harder problem than deploying AI inside one.

(Adapted from FirstPost.com)

Categories: Creativity, Economy & Finance, Regulations & Legal, Strategy