Zuckerberg's Two Faces on AI Progress

Here is a test of your credulity. On the same day last week, Meta offered the world two competing versions of its AI progress. First, from Alexandr Wang, the company's AI leader — a man whose job is to project confidence — came the announcement that Meta's upcoming model, code-named 'Watermelon,' has achieved performance parity with OpenAI's GPT-5.5. No independent verification, naturally; just the kind of internal cheerleading that usually precedes a product launch or a funding round. Then, from Mark Zuckerberg himself, during a staff Q&A, came a far less tidy admission: AI agents haven't 'accelerated in the way' that company executives had anticipated. Which Meta are we meant to believe?

Take Wang's claim first. Parity with GPT-5.5 is a bold statement, especially from a company that until recently was seen as a follower in the large language model race. But even if it were true — and the lack of any external validation should give us pause — it reveals something about the industry's priorities. Meta, like every other tech giant, is obsessed with benchmark-beating as a proxy for real-world utility. Yet benchmarks are to AI what exam results are to a politician: a measure of performance under artificial conditions, not a guarantee of competence in the messy world beyond the test centre. Wang's Watermelon may well match GPT-5.5 on standard evals; whether it can book a meeting or file an expense report without hallucinating is another matter.

Which brings us to Zuckerberg's reluctant candour. AI agents — the holy grail of autonomous assistants that can handle complex tasks from start to finish — have not progressed as hoped. This is the sort of admission that rarely emerges from a tech CEO's mouth unprompted. That Zuckerberg felt compelled to share it with employees suggests the gap between internal expectations and external marketing has become too wide to ignore. The contrast with Wang's buoyant confidence is jarring. Either the CEO is out of touch with his own AI division, or the AI leader is putting a brave face on a stalled engine. Neither possibility is reassuring.

This two-faced storytelling is not unique to Meta. It is a symptom of an industry that has built a towering narrative on a foundation of vapour. Every quarter, companies announce breakthroughs; every year, the promised revolution of autonomous agents and fully automated workflows recedes further into the future. The result is a cognitive dissonance that investors and journalists have been reluctant to name. We are told that AI is advancing at an exponential rate, yet here is the CEO of a major player admitting, in the privacy of an internal meeting, that progress is slower than anticipated. Which message wins? The one that sells stock, presumably.

The implications for the labour market are worth noting. Last week, I wrote about Ford rehiring 350 veteran engineers after AI-powered cameras failed to capture their tacit knowledge. That story and Zuckerberg's admission are two sides of the same coin: the belief that AI can replace human expertise quickly is a fantasy. If even Meta's own agents are moving slowly, what does that say about the predictions of mass job displacement that have dominated headlines? The 'skills gap' narrative — that workers must retrain for an AI-dominated future — looks increasingly like a convenient excuse for cost-cutting, not a description of reality.

None of this means AI is worthless. It means the hype machine has overshot. Companies like Meta are caught between the need to sustain investor enthusiasm and the mundane truth that building reliable, autonomous systems is hard. They sell us a future that is always just around the corner, then hide their frustration when the corner recedes. Zuckerberg's honesty is refreshing, but it also confirms what many have suspected: the emperor's new clothes are a little threadbare. The question is how long the market will pretend it can't see through them.