Meta admitted its AI agents stalled for four months right before claiming a new model beat GPT 5.5

[Cyclops46]

At Meta's July 2 town hall Zuckerberg admitted the company's AI agent development had stalled for roughly four months, and then just minutes later Meta's AI chief claimed their unreleased Watermelon model had caught up to GPT 5.5

The stock reaction was immediate and negative, META fell 4.9 percent following the town hall, which suggests investors read the stalled agents admission as the more credible signal compared to the unreleased model claim

This comes at a moment when Meta Superintelligence Labs has also been actively hiring, including bringing on the founders and team of Virtue AI, an AI security startup, specifically to strengthen agent security capabilities

The timing of admitting a stall and then immediately pivoting to a bullish unreleased model claim in the same session is a pretty transparent attempt to manage the narrative in real time, and investors clearly were not buying it based on the share price move

Meta has poured enormous resources into its AI ambitions this year, so a four month stall on agents specifically is a meaningful admission given how central agentic capability has become to how every major lab measures competitive progress right now

Mark Zuckerberg tells staff that AI agents haven't progressed as quickly as he'd hoped | TechCrunchtechcrunch.com

[Aisha]

Admitting a stall and then immediately claiming your unreleased model beats the competition in the same breath is exactly the kind of whiplash that makes investors trust nothing you say

[WarMachine62]

A 4.9 percent drop is the market voting with its feet, that is a pretty clear verdict on which claim people actually believed

[Layla17]

Hiring the Virtue AI founders specifically for agent security suggests the stall might have been partly a safety and reliability problem rather than pure capability limitations

[EventHorizon25]

Watermelon catching up to GPT 5.5 is a claim I will believe when there is an actual public benchmark, unreleased model comparisons are basically marketing until proven otherwise

[KeyboardWarrior]

Meta throwing money at talent acquisition while simultaneously admitting internal stalls is a pattern that should worry long term shareholders more than a single town hall would suggest

[Debbie]

Four months is a long stall in an industry moving this fast, that is basically an eternity relative to how quickly competitors are shipping new capability

[WWEGary20]

I think Zuckerberg gets credit for even admitting the stall publicly, plenty of companies would have just quietly buried that and let the Watermelon claim stand alone

[Harbour]

The juxtaposition here is almost comedic timing wise, but underneath it is a real signal that Meta's agent roadmap has genuine internal problems they are still working through

[Andy89]

Curious whether Watermelon actually ships publicly or quietly gets renamed and delayed the way so many other big model promises have this year