On August 26, 2026, an in-depth Reuters investigative report uncovered secrets behind Meta’s internal initiative codenamed Project OT (Organization Transformation): The tech giant pouring hundreds of billions of US dollars annually into AI once planned to deploy AI Agents to take over a large volume of employees’ work. Some teams drafted plans to cut workforce by up to 60%. In the end, this radical transformation project to build an “AI-native company” collapsed entirely.

On August 26, 2026, an in-depth investigative report by Reuters uncovered the secrets behind Meta’s internal initiative codenamed Project OT (Organization Transformation):

The tech giant investing hundreds of billions of US dollars annually in AI once planned to deploy AI Agents to take over a large portion of employees’ workloads. Some teams drafted plans to cut their workforce by up to 60%. Ultimately, this radical transformation project to become an “AI-native company” collapsed completely.

The grand vision finalized by Mark Zuckerberg during a closed-door executive meeting in Hawaii aimed to build an entirely new corporate structure:

A small group of elite employees would oversee operations, while autonomous AI Agents handled most day-to-day development, maintenance and procedural work. This was intended to slash labour costs and amplify AI output, with organizational restructuring rolled out in two phases: the first round of layoffs in May, followed by larger-scale optimisation in November.

Reality shattered this ambitious blueprint.

Hours before the first round of layoffs was set to launch, Zuckerberg urgently halted the planned second wave of large-scale redundancies scheduled for November. Only the first round of roughly 8,000 layoffs was carried out, accounting for 10% of the total workforce. Meanwhile, 6,000 open roles were closed, and 7,000 employees were reassigned to AI-related departments. The high-profile scheme to replace staff with AI was officially terminated.

Beneath the lofty vision lie three revealing sets of internal data

Figures disclosed in Meta’s internal documents clearly explain the underlying reasons for Project OT’s failure: AI generates massive volumes of content, yet delivers severely insufficient valid output alongside substantial risks.

1. Code changes surged by 220% year-on-year, but user-facing new feature releases rose by merely 36%.

AI Agents churned out code in bulk, yet the vast majority amounted to useless "digital noise" that could not be turned into genuine product iterations delivered to end users.

Engineers had to review, revise and debug the flood of AI-generated content one by one. Instead of reducing workloads, this created an enormous new screening burden.

2. Severe technical and security incidents climbed 40% year-on-year, and engineers spent 70% more time firefighting faults.

Internal Meta memos issued stark warnings: AI Agents operating without human guardrails can carry out "large-scale destructive actions that humans would almost never perform".

Multiple high-severity security incidents occurred over the course of the project: AI Agents provided incorrect technical recommendations, leading to sensitive data exposure that lasted two hours. Hackers exploited vulnerabilities in AI customer service bots to hijack high-profile accounts, including Barack Obama’s official White House Instagram account, sparking panic inside the company.

Rather than taking over work, AI created countless bugs and risks for existing staff to remediate. Many engineers reported their core focus shifted from product development to fixing errors produced by AI.

3. Employee satisfaction plummeted from 74% to 55%, and internal protests erupted repeatedly.

To train its AI Agents, Meta deployed keyboard and mouse tracking tools to record employee operational behaviours for model training.

Once workers realised the firm was training AI to replace their roles, internal forums were flooded with complaints. Large numbers of core engineers began submitting job applications externally, talent flight risk spiked sharply, and team morale suffered heavy damage.

At an all-hands meeting in July, Zuckerberg was forced to acknowledge reality: over at least the prior four months, AI Agent development had failed to meet company expectations.

Meta’s official public statement framed Project OT merely as one of multiple scenario simulations, not a fully confirmed rollout plan.

However, Reuters’ interviews with more than 20 insiders and reviews of extensive internal documents and recordings confirmed that management did treat large-scale AI workforce replacement as its core reform direction, only abandoning the radical roadmap due to technical limitations.

Important distinction: Meta is not abandoning AI tools. The company continues to pour hundreds of billions into AI infrastructure and still encourages employees to use AI to assist coding and boost productivity. What was halted was the aggressive strategy of relying directly on autonomous AI Agents to replace jobs at scale.

Three core reasons behind Project OT’s failure

1. Agent technology is far from ready for autonomous production-grade work

Current AI excels at discrete single tasks: writing a snippet of code, drafting copy, or offering recommendations. But real enterprise systems are highly complex, burdened by years of technical debt, with countless business rules locked inside veteran employees’ experience.

Autonomous AI Agents lack holistic business judgement and risk awareness. They tend to produce seemingly reasonable yet devastating operations. They can complete fragmented tasks but cannot independently take end-to-end accountability for a full job function.

AI can act as an assistant, but it cannot yet function as a fully independent employee.

2. The “AI replaces humans” narrative erodes organizational trust

When a company sets “training AI to replace existing staff” as an internal objective, employees naturally become defensive. Top talent will prioritise leaving, and those who stay suffer drastically reduced job security. Technical transformation cannot come at the cost of team trust.

AI tools should empower employees instead of directly targeting their replacement. Getting this priority backwards triggers brain drain and ultimately undermines the firm’s long-term competitiveness.

3. The efficiency trap: output quantity ≠ business value

Meta fell into a common pitfall seen in many corporate AI transformations: treating lines of AI-generated code and document volume as efficiency metrics.

AI can easily inflate process metrics, yet real value lies in delivering stable, usable products that solve user needs.

Chasing only “how much work gets done” while ignoring “how much work succeeds” generates piles of useless deliverables and technical debt, which ultimately increases manual workload.

Practical takeaways for international students and professionals: what skills are hard to replace in the AI era

Meta’s costly internal trial delivers key lessons for overseas students hunting jobs and professionals worldwide.

Claims that “AI will wipe out huge numbers of jobs” have circulated widely in recent years, yet Meta’s real-world experience shows AI reshapes tasks rather than simply eliminating established roles in bulk.

1. AI handles standardized execution well, but struggles with complex, holistic decision-making

Standardised tasks such as writing code snippets, compiling materials and basic copywriting will be heavily automated by AI. However, understanding complex business context, weighing multi-faceted risks, handling edge-case anomalies and cross-team decision-making remain heavily human-led. During job hunting, do not only build narrow technical skills; intentionally cultivate big-picture thinking, risk assessment and complex problem decomposition capabilities.

2. Human-AI collaboration is a core future competency, not a reason to reject AI

Do not panic over AI, nor blindly trust it. Workplaces increasingly need people who can steer AI, validate its outputs and correct its mistakes. Meta’s engineers were not eliminated by AI; instead, they were forced to spend massive time fixing AI errors. The ability to judge AI output quality and leverage AI as a productivity amplifier will separate top performers from others.

3. Technical roles are not “more vulnerable if they are more entry-level”. Deep industry expertise, knowledge of legacy systems and cross-scenario problem-solving experience are hard to replicate in the short term. The value of many senior staff lies in their awareness of historical system constraints and pitfalls — tacit knowledge that current AI struggles to fully learn.

4. When looking at big tech layoffs, distinguish between two scenarios: layoffs for financial optimisation, and AI-driven job replacement.

Meta’s latest redundancies were partly cost-cutting and partly business realignment; large-scale AI replacement of staff never materialised. Avoid being swept up by online anxiety, and view AI’s impact on the job market objectively.

Closing thoughts

The failure of Project OT is not a failure of AI itself, but the collapse of an over-optimistic commercial fantasy about AI capabilities.

AI tools undoubtedly hold value, but at this stage, they work best as collaborative human assistants rather than “digital labour” deployed to replace employees in bulk.

AI will change job responsibilities, yet fully taking over complete job functions remains far from reality.

For job seekers, there is no need to live in constant fear of being replaced by AI. Keep iterating your skill set:

Treat AI as a tool, and build your moat around complex decision-making, business insight and interpersonal collaboration. This is a more sustainable strategy for the AI era.