Instagram head Adam Mosseri says Meta will probably have to put hard limits on how many AI tokens each engineer can spend, and he thinks that day is a year or two out. He said it on Lenny’s Podcast, according to TechCrunch AI, and the number he floated should stop you cold: a strong engineer’s token burn could match their entire cost of employment.
Read that again. The AI bill for one developer, roughly equal to their salary.
What Mosseri actually said
“I think that you can imagine, at least in a year or two … that the burn rate of a strong engineer might be the same as their salary, or their cost of employment. And in that world, you’re going to probably need to put in some caps,” Mosseri said.
His framing is boring on purpose. Tokens are just another line item. “I think of it like…any other resource,” he said, putting them next to GPUs, CPUs, storage, RAM, labeling budgets and payroll. All things he already has to divide up across teams every quarter.
One detail worth holding onto: Meta has no token caps for anyone right now. Mosseri thinks they’d be healthy later, and that each engineer’s cap would scale with how much the company trusts them to spend in an “ROI-positive” way.
This is bigger than Meta
TechCrunch AI lays out the pattern, and it’s not one company panicking:
- Meta killed an internal AI token spend leaderboard after AI costs put the company on track for billions of dollars in 2026.
- Uber burned through its entire 2026 AI coding budget by April.
- Microsoft cancelled Claude Code licenses and pushed engineers onto its own Copilot CLI instead.
The status quo two years ago was unlimited AI as a recruiting perk. Give everyone the best tools, don’t count the pennies, move fast. That era is closing. Finance has walked into the room.
Why this matters
The leaderboard story is the most revealing part. Meta built a scoreboard for token spend, engineers competed on it, and the bill exploded. “It’s not that hard to build a token incinerator, and that doesn’t create a lot of value,” Mosseri said.
What stands out here is the shift in how AI usage gets judged. Usage was the metric. More prompts meant more adoption meant progress. Mosseri is quietly retiring that idea and replacing it with output per dollar. That’s a different question, and most teams have no way to answer it yet.
There’s a second signal underneath. If a single engineer can spend their own salary on inference, the cost structure of software has changed shape. Headcount used to be the whole story. Now compute sits right next to it on the same budget line.
The next two years
Here’s where this lands:
- Token budgets become a normal management tool. Managers will allocate them the way they allocate headcount. Expect dashboards, quotas and approval flows for the big jobs.
- Trust becomes a currency. Mosseri said caps would be proportional to trust. Engineers who show clean returns on their spend get more room. That’s a new kind of performance review.
- Prices fall, but not fast enough to save you. Mosseri expects model makers to enter a pricing war for users. He’s probably right. He also expects caps anyway, because consumption grows faster than prices drop.
- The “silly things” get cut first. Meta already trimmed costs just by shutting down its own gamified waste. Most companies have their own version sitting somewhere.
What to do now
Start measuring. If you can’t say which AI workflows actually pay for themselves at your company, you’ll find out the hard way when someone above you sets the cap without your input. Teams that already track cost per shipped feature will negotiate from data. Everyone else will negotiate from vibes.
The cheap experimentation window is closing. Use it while it’s open, and know what your spend is buying. More details are in the original TechCrunch AI report.