Meta scrambles to refocus its resources on usable AI products and features after a decade of research in its AI division disincentivized work on generative AI
it's also about how gen AI fever can seize hold of one of the world's largest giants. w/ @sal19 & @dseetharaman https://www.wsj.com/...
Context & Ripple Effects
This story closes out a months-long arc of Meta playing catch-up: earlier reporting detailed how the company was slow to adopt pricey AI-optimized systems and scrapped a custom AI chip, and how its research-first culture left generative work under-resourced even as rivals shipped. The same month, Google was running its own scramble to infuse generative AI into every product — evidence that the ChatGPT shock hit every large incumbent's org chart at once.
What makes this piece durable is what it reveals about incentives: a decade of rewarding publications over products meant the division structurally deprioritized generative AI, and management is now forcibly rewiring that. The follow-through shows in later moves — splitting up the Responsible AI team into the generative unit and an infrastructure group, and eventually a division-wide hiring freeze gated by Chief AI Officer Alexandr Wang as priorities tightened further.
First-order effects
- Meta's AI researchers and product teams are being redeployed from open-ended research to shippable generative features across Meta's apps, with internal incentives rewritten to favor usable output over publications.
- The generative AI team formed in February 2023 becomes the destination for resources and staff — including most of the former Responsible AI group — consolidating decision-making in one product-oriented unit.
Second-order effects
- Google and other incumbents face the same forced choice between research prestige and product velocity, and Meta's visible whiplash gives every big-lab leader cover to justify similar reorganizations internally.
- Meta's late adoption of expensive AI-optimized hardware and its abandoned custom chip leave it dependent on external compute at exactly the moment it needs to ship fast, raising the cost of the catch-up.
Third-order effects
- If the pattern holds, the big-company AI lab as an autonomous research fiefdom gives way to AI divisions run as product organizations — with safety and responsibility teams folded into shipping units rather than standing apart, a structural trade-off regulators will eventually have to weigh.
- The episode argues that distribution-rich giants can absorb a slow start by redirecting capital and headcount at scale — but only if their internal incentive systems flip faster than the competitive cycle demands.
The trend: Big Tech AI divisions are being restructured from research labs into product-shipment machines, with incumbent scale substituting for first-mover speed.