Remember when a $20-a-month “Pro” or “Plus” AI plan served up more AI access than you could possibly use? Ah, those were the days.
I’ve been dipping in and out of ChatGPT Pro, Google AI Pro, and Claude Pro for at least a couple of years now, and up until five months ago, I never, ever hit a usage limit.
Like many everyday AI users, I was doing little more than conversing with ChatGPT, Gemini, and Claude in a humble chatbox, asking random questions about recipes and resumes, or occasionally pasting an error log from my Raspberry Pi. You know, really basic stuff.
Everything changed for me last December, when I got my first taste of Google Antigravity, one of the first “agent-first” AI coding assistants. I’d previously heard of Cursor and Claude Code but never tried them, but the second I saw Google Antigravity dive into my Raspberry Pi and code on its own, making quick work of software problems that had vexed me for months, I was hooked.
Looking back, Google Antigravity was a pivotal “I know kung fu!” moment in my AI journey, and it also marked the first time I hit a usage limit. At first, I thought the arcane error code in the Antigravity chat sidebar meant I’d smacked into some kind of server error. But like a toddler getting their first ouchie on a hot stove, I soon realized what really happened: I’d blown through a multi-hour Gemini usage quota in a matter of minutes.
Fast-forward to today, as I (and likely plenty of other AI users) hopscotch between Antigravity, Claude Code, and OpenAI’s codex, juggling $20-a-month plans for each service and squeezing as much agentic AI coding in as I can without hitting those damnable limits.
Of course, the big AI providers are responding in kind, frantically adjusting and readjusting the usage limits on their flat-rate plans–sometimes invisibly–as hoards of everyday AI users unleash armies of AI agents on their desktops.
Anthropic has probably drawn the most ire for tweaking its flat-rate plans, quietly dialing down the usage allowances for Pro ($20 a month) and Max ($100 and up) users to keep up with the agentic AI demands of Claude Code and Claude Cowork, while gradually transitioning Enterprise subscribers to a usage-based payment model.
Google and OpenAI have also been busy adjusting their respective rate limits as Antigravity and Codex usage takes off, with OpenAI recently splitting its Pro tier into $100-a-month and $200/month plans, thus creating a middle ground for $20/month Plus users looking to level up.
Most recently, GitHub has taken the extreme course of halting all new signups for its Copilot Pro and Pro+ plans ($10 a month and $39/month, respectively). GitHub also announced that starting now, it will tighten usage limits for individual users while yanking access to Opus 4.7, the most powerful Claude model (save for the “too-powerful-for-you” Claude Mythos) from the Pro tier.
GitHub, which said it will offer full refunds to subscribers who want to cancel due to the sudden changes, didn’t mince words when explaining its new policies:
Agentic workflows have fundamentally changed Copilot’s compute demands. Long-running, parallelized sessions now regularly consume far more resources than the original plan structure was built to support. As Copilot’s agentic capabilities have expanded rapidly, agents are doing more work, and more customers are hitting usage limits designed to maintain service reliability. Without further action, service quality degrades for everyone.
The real problem, of course, is that AI providers from Anthropic and Google to OpenAI and GitHub have been unleashing more and more powerful agentic AI tools on flat-rate users with (apparently) little regard to whether they actually have the capacity to support them. That leads to the big AI players continually messing with usage limits while flat-rate subscribers struggle to keep up.
It’s a bad experience for everyone, and I don’t see it improving anytime soon.
