exhausted their weekly quotas after just an hour of work
When a tool becomes load-bearing infrastructure for daily work, even a brief restriction feels like a wall appearing mid-stride. This week, Google discovered that truth firsthand: usage limits introduced for Gemini's Antigravity coding assistant were exhausted by developers within a single hour, prompting two rounds of tripling in a matter of days. The episode is less a story about quotas and more a story about the gap between how platform architects imagine usage and how builders actually live inside their tools.
- Developers hit their weekly Gemini quotas inside a single hour of coding work, turning a productivity tool into a source of sudden, disorienting friction.
- The backlash was swift and loud enough that Google tripled Antigravity's rate limits on Wednesday and reset quotas for all paid users — only for the new ceilings to fall again by evening.
- A second tripling of weekly quotas followed that same night, with Google's own DeepMind director acknowledging that a couple of work sessions could drain an entire week's allowance.
- Even after both increases, limits remain below what developers had before restrictions arrived, leaving the underlying tension between infrastructure costs and real-world usage unresolved.
- The rest of Gemini's toolset still operates under the original tighter constraints, signaling that Antigravity's relief may be a preview of broader friction yet to come across the ecosystem.
Google introduced compute-based usage limits for Gemini this week as part of announcements at Google I/O, framing them as a way to manage demand and ensure fair access. What the company did not anticipate was how quickly developers working inside Antigravity, its AI-powered coding assistant, would exhaust those allowances — some burning through their entire weekly quota after barely an hour at the keyboard.
The response came fast. On Wednesday, Google tripled Antigravity's rate limits and reset weekly quotas for all paid plans, a clear admission that the initial caps had been miscalibrated. But the fix didn't hold. By that evening, users were hitting the new ceilings too.
Varin Mohan, a director at DeepMind overseeing Antigravity, acknowledged the problem openly: a few ordinary work sessions were enough to drain a week's worth of access. That same night, Google tripled the weekly quota a second time and issued another reset across all paid tiers.
The rapid back-and-forth exposes a genuine tension — between the cost of running large-scale AI infrastructure and the reality that developers have woven these tools into the fabric of their daily work. Even with both increases applied, limits remain lower than they were before restrictions arrived. And Antigravity's adjustments are isolated; the broader Gemini suite still operates under the tighter original constraints, suggesting that other developers across the ecosystem may encounter the same wall in the weeks ahead.
Google introduced usage limits for Gemini this week, and within hours, developers working in Antigravity—the company's AI-powered coding assistant—were slamming into those caps. The frustration was immediate and vocal enough that Google responded twice in a single week, each time tripling the allowances.
The limits arrived as part of a broader set of announcements at Google I/O, where the company outlined compute-based restrictions across Gemini's various tools and capabilities. The idea was to manage demand and ensure fair access. What happened instead was that some users exhausted their weekly quotas after just an hour of work in Antigravity, a jarring shift from the previous, more permissive regime.
On Wednesday, Google moved quickly. The company tripled Gemini's rate limits specifically for Antigravity and reset the weekly quotas for every user on paid plans. It was a clear signal that the initial caps had been set too low. But the relief was short-lived. By that same evening, users were again bumping against the new ceilings.
Varun Mohan, a director at DeepMind overseeing Antigravity, acknowledged the problem directly. Users could exhaust their weekly allowance after just a couple of work sessions, he said. That same night, Google tripled the weekly quota a second time and reset quotas across all paid tiers once more.
The back-and-forth reveals a company caught between two pressures: the need to manage infrastructure costs and the reality that developers depend on these tools for their actual work. Even with the increases, though, the limits remain lower than they were before Google introduced restrictions. Mohan's team has adjusted only Antigravity's quotas; the rest of Gemini's suite—other tools and abilities—still operates under the tighter constraints that prompted the original complaints. That asymmetry suggests Google is still working out where the real ceiling should be, and that developers in other parts of the Gemini ecosystem may face similar friction in the weeks ahead.
Notable Quotes
Users could hit their weekly limits after a couple work sessions— Varun Mohan, Director at DeepMind overseeing Antigravity
The Hearth Conversation Another angle on the story
Why did Google set the limits so low in the first place if they knew developers would hit them in an hour?
That's the question everyone's asking. Compute costs are real, and Google probably wanted to be conservative—see what actual usage looked like before opening the floodgates. But they underestimated how intensively people use Antigravity.
So they're just guessing at the right number?
Essentially, yes. They're adjusting in real time based on feedback. The fact that they tripled twice in one week tells you they're still calibrating.
Why only fix Antigravity and not the other Gemini tools?
Antigravity is the squeaky wheel. Developers use it constantly for their actual jobs, so the pain is immediate and visible. Other tools might have lighter usage patterns, or the users affected are less vocal.
Are the new limits actually enough now?
Probably not permanently. They're higher, but still below where they started. This feels like a temporary patch, not a final answer.
What does this say about Google's planning?
That they're reactive rather than proactive. They should have modeled this before launch. Instead, they're learning from users hitting walls.