Meta is currently one of the companies investing the most in artificial intelligence, but even the scale of its investment does not guarantee quick results. Mark Zuckerberg admitted to staff that the development of AI agents is progressing more slowly than management had anticipated, which shows that there is still a clear gap between ambitious plans and their implementation.
During an internal meeting with staff, Meta’s chief executive said that the company’s reorganisation, carried out in the first half of the year, had not yet delivered the expected results. The company made around 10 per cent of its global workforce redundant and transferred nearly 7,000 employees to teams working on artificial intelligence. The aim was to accelerate work on AI agents and to fund growing investments in infrastructure.
As Zuckerberg admitted, the restructuring “wasn’t as clean as it could have been”, and management had misjudged the pace of change. However, the most important factor turned out to be the assessment of the agent technologies themselves.
“The trajectory of agent development over at least the last four months has not, in fact, accelerated in the way we expected”
This refers to AI systems capable of independently performing complex tasks on behalf of the user. As recently as the start of the year, Meta’s management had assumed that progress in this field would be much faster, thanks in part to new development tools such as Claude Code from Anthropic.
Despite the slower pace of development, Meta is not changing its strategy. The company plans to invest up to $145 billion in AI infrastructure this year, remaining one of the world’s largest investors in this field. Zuckerberg emphasised that he expects more visible results from these investments within the next three to six months.
During the same meeting, Meta also addressed the controversy surrounding software used to monitor staff activity in AI projects. Chief Technology Officer Andrew Bosworth reported that the completed review found no evidence of employee data being used to train AI models. If the programme is relaunched, it will operate solely on the basis of users’ voluntary consent.
“For people who feel comfortable with it, that’s great – they can contribute to this sort of wonderful human survey. For people who don’t, it’s not a problem.”
