Mark Zuckerberg Unhappy With AI Agent Delays at Meta
Meta’s AI strategy currently presents a split picture. In an internal meeting, the recording of which was reviewed by Reuters, CEO Mark Zuckerberg stated that the development of “agentic” technologies—AI systems capable of performing tasks autonomously—has not accelerated as expected over at least the last four months.
Zuckerberg admitted that the hope for rapid, widespread adoption of these agents has not yet materialized. This comes at a time when Meta has already aggressively redirected resources into the corresponding infrastructure. The impact of this strategic realignment was already evident in May, when the company cut approximately 10% of its workforce and restructured around 7,000 employees into AI-focused teams—a process Zuckerberg acknowledged was not as clean as it could have been due to miscalculations regarding timing.
Expansion Despite Delays
Despite the cautious assessment regarding development speed, Meta is pushing the application of the technology forward. On Thursday, the company expanded the “Meta Business Agent” globally. The feature, which can be used on Instagram, Messenger, and WhatsApp, is intended to allow businesses to answer customer inquiries, make product recommendations, and close sales without human intervention. Zuckerberg expressed optimism that massive AI investments will begin to pay off within the next three to six months.
Technological Progress in Models
An opposing, optimistic signal came from the company’s technical leadership. Alexandr Wang, head of the newly created “Meta Superintelligence Labs,” told employees that Meta’s upcoming AI model, codenamed “Watermelon,” has already caught up with OpenAI’s flagship model, GPT-5.5, in benchmarks.
“Watermelon,” which builds upon the “Avocado” model (internally known as Muse Spark), is said to use an order of magnitude more computing power than its predecessors. While Meta has previously struggled to convince developers of its models’ leading-edge position, this progress could mark a turning point. Wang also indicated that updates for the current Muse Spark model will follow soon, specifically aimed at improving coding and agentic capabilities.
Massive Investments and Economic Context
This technological push is backed by enormous financial resources. Meta has revised its spending forecasts for chips, data centers, and infrastructure upward. For the current year, the company expects investments between 125 and 145 billion USD.
This expansion is taking place in a market environment where the convergence of AI and blockchain technology is considered a major growth market. While leaders in the crypto industry predict that AI agents will become the primary users of blockchain payments, current data shows a still-nascent market: trading volume via the AI-agent-supported x402 protocol was only approximately 2 million USD over the past 30 days.
Meta thus remains in a state of tension: while technical hardware and model performance appear to be growing rapidly, the practical application of autonomous agents in everyday user life is still waiting for the decisive breakthrough.
