Meta was considering a two-stage reorganisation of its workforce to an «AI-native» model, under which individual teams could be reduced by up to 60%. Following the redundancy of 10% employees on 20 May, Mark Zuckerberg cancelled the second wave of redundancies, which had been scheduled for November. The decision was taken against a backdrop of open resistance from staff and internal indications that AI agents had not delivered the expected productivity gains.
Briefly about the main points
- Project OT was intended to automate part of the work carried out by thousands of staff members.
- The scenarios involved redundancies affecting specific teams, rather than Meta’s entire workforce.
- The restructuring in May involved the redundancy of 10% employees.
- Internal surveys have revealed a sharp decline in staff morale.
- Meta remains under financial pressure due to high expenditure on AI infrastructure.
Project OT envisaged two waves of restructuring
The plan, codenamed Project OT Meta’s management team was formed in January at the annual meeting in Hawaii. The aim was to make the company «AI-native»: digital workers would carry out a significant proportion of day-to-day tasks, whilst smaller, talent-rich teams would oversee their work.
The scenarios envisaged waves of restructuring in May and November. They involved redundancies, transfers to new departments, the closure of vacant posts and the dismissal of those the company considered to be underperforming employees. One of the HR managers estimated that the scale of the redundancies would be no less than approximately a quarter of the workforce, which Meta cut in late 2022 – early 2023.
Meta confirmed to Reuters the existence of the project and the two-phase approach. The company acknowledged that the most stringent models allowed for a reduction in the size of some teams to 60%, but emphasised that this was not a plan to make 60% of all staff redundant. According to the company, the final number of redundancies had not been determined until the second phase was cancelled.
Smaller teams, faster cycles and the search for «irreplaceable» specialists
In practice, the model was tested through small technical iterations: two or three engineers and a designer were tasked with creating prototypes in four-week cycles, rather than adhering to traditional six-month product release schedules. By June, at least 11 engineering and research departments had already adopted this approach.
Under the new structure, the traditional roles of designers and engineers could be replaced by the all-round role of «builder». A head of a major division would be responsible for 30–50 people, whilst team leaders would coordinate day-to-day work without formal managerial authority. In one division, performance appraisals and promotions were to be determined by division heads, with support from HR and unnamed AI systems. Meta stated that decisions on appraisals and promotions are made by people, not artificial intelligence.
The company has also launched an HR tool to identify «irreplaceable talent» — in particular, specialists at the so-called «10X Performer» level. Part of the funds saved through redundancies, Meta planned to allocate funds towards generous compensation packages to recruit and retain top AI engineers.
Rumours of job cuts have intensified the conflict within the company
Reuters reports in March and April regarding potential staffing decisions caused concern amongst rank-and-file staff even before many vice-presidents had been briefed on Project OT. Management chose not to discuss these reports publicly with staff, whilst senior managers were advised to explain to their teams that their roles would «evolve» as a result of AI.
Further outrage was caused by the installation on staff members’ devices in the US of software that recorded keystrokes and mouse movements for training purposes AI agents interact with a computer. On Workplace, staff wrote that they could train the technology that would eventually replace them; in their replies to managers, they posted images of elephants as a symbol of the obvious but unspoken problem of redundancies.
By the end of May, the strength of some engineering units had fallen by 30% due to transfers and dismissals. The proportion of positive responses in the internal Pulse survey fell from 74% to 55%, whilst attempts at worker self-organisation intensified.
The increase in the volume of code was accompanied by incidents
Meta’s internal data showed that the wider use of AI had led to a sharp increase in the volume of code generated, but had not yielded a commensurate return for the products. According to figures cited by Chief Technology Officer Andrew Bosworth in early June, changes to internal platforms and infrastructure rose by 220% year-on-year, whilst changes that added or updated features for users increased by 36%.
As early as March, infrastructure teams had warned of signs of reliability issues due to the influx of AI code. Another internal memo stated that uncontrolled agents were carrying out large-scale destructive actions that were atypical of humans. Major technical and security incidents rose by 40% compared with the previous year, whilst the time taken to resolve them increased by 70%.
In early June, cybercriminals exploited Meta’s new AI support bot to gain access to high-profile Instagram accounts, including the inactive Obama White House page. The company did not comment on internal metrics regarding the glitches, but stated that data generated by its new Applied AI Engineering division had helped train the AI model released last month.
Zuckerberg admitted to a mistake regarding the timing of the reorganisation
A few hours before the May redundancies, Zuckerberg discussed the situation with his immediate senior management and put the planning for the November phase on hold. Following the redundancies, he wrote to staff saying he did not expect any further company-wide redundancies this year and wanted to provide the team with greater stability. Meta suspended its mouse-tracking programme, allowed some of the redeployed staff to return to their previous teams, and sought to boost morale through additional office benefits.
In July, at an internal meeting, Zuckerberg acknowledged that the reorganisation had not gone to plan: AI agent technology was developing more slowly than he had anticipated. At the same time, he expected to see more tangible benefits over the next three to six months. The pressure to find savings remains: this year, Meta plans to invest at least $130 billion in AI chips and other infrastructure, which, according to estimates by LSEG analysts, will absorb its operating cash flow in 2026.
