Earlier this year, Meta, one of the industry’s leading advocates for AI, was poised to significantly reduce the size of some teams and replace workers with AI as part of Project OT (Organization Transformation). The plan aimed to make Meta “AI native,” but internal data and employee pushback ultimately led the company to back away from the initiative.
According to a Reuters investigation published on Wednesday, Meta’s internal data indicated that the planned overhaul was not achieving the desired results. For instance, Meta CTO Andrew Bosworth reported in early June that code changes to internal software platforms and infrastructure increased by 220% year over year. However, changes that led to new or upgraded features reaching Meta users only rose by 36%. This disparity highlighted the inefficiencies introduced by the AI driven changes.
Furthermore, Meta executives identified “reliability warning signs” caused by the AI coding surge. An internal post in April noted that “unchecked AI agents were performing large scale, disruptive actions that humans are unlikely to execute.” As a result, major technical and security incidents, such as service disruptions and possible data leaks, spiked 40% from the previous year. The time staffers had to spend firefighting these incidents increased by 70%.
In April, Meta mandated that tracking software be installed on U.S. employees’ devices to capture their keystrokes and mouse clicks, intending to teach AI agents to mimic human interactions with computers. However, this move backfired, with employees rebelling due to concerns that they might be training their own AI replacements.
To try and assuage worker concerns, Meta promised to increase spending on travel and social events and improve snack quality in office microkitchens. However, these measures did not significantly boost morale. According to Reuters, Meta CEO Mark Zuckerberg continued to use terms like “company wide” and “this year” in internal communications about layoffs, prompting speculation that he might proceed with team specific cuts or performance based dismissals or delay company wide reductions until next year.
Consultants and analysts emphasized the importance of careful AI implementation, drawing parallels between Meta’s experience and broader industry challenges. Sanchit Vir Gogia, chief analyst at Greyhound Research, cautioned that Meta trusted an AI forecast before it existed in production, a different failure from over reliance on AI. He noted, “Prove the action before widening the authority, and the authority before removing human control. Only then is removing human capacity a decision, not a bet.”
Terra Higginson, a principal research director at Info Tech Research Group, argued that unchecked AI agents and human displacement are not desirable outcomes. “We want work to be reimagined so the human part still matters and technology makes it better,” she said. She pointed out that AI can generate a lot of output without always delivering the desired outcomes. “We should not use AI output as a proxy for productivity,” she added.
Tom Findling, CEO of Conifers.ai, suggested that IT leaders should use Meta’s experience as an opportunity to argue for a more thoughtful AI rollout. “Say that you may not get a 500% productivity boost, but IT can show them a meaningful way to get 300%,” he advised. Findling highlighted the need to focus on measurable business outcomes rather than just the volume of work generated by AI.
Justin Greis, CEO of Acceligence, added that the key takeaway from Meta’s experience is the disconnect between AI activity and actual value creation. “AI can make an organization extraordinarily busy without necessarily making it more productive,” he said. “We have spent decades teaching technology leaders that lines of code, tickets closed, and projects launched are imperfect proxies for business value. AI makes that measurement problem much more acute.”
Greis emphasized that the most critical question for executives should be, “What measurable business outcome improved because AI produced it?” His advice underscores the need for a more nuanced and strategic approach to AI implementation in the enterprise.