Flat-rate AI plans are damaged. Blame AI brokers

Bear in mind when a $20-a-month “Professional” or “Plus” AI plan served up extra AI entry than you may presumably use? Ah, these had been the times.

I’ve been dipping out and in of ChatGPT Professional, Google AI Professional, and Claude Professional for at the very least a few years now, and up till 5 months in the past, I by no means, ever hit a utilization restrict.

Like many on a regular basis AI customers, I used to be doing little greater than conversing with ChatGPT, Gemini, and Claude in a humble chatbox, asking random questions on recipes and resumes, or sometimes pasting an error log from my Raspberry Pi. You realize, actually primary stuff.

Every part modified for me final December, after I obtained my first style of Google Antigravity, one of many first “agent-first” AI coding assistants. I’d beforehand heard of Cursor and Claude Code however by no means tried them, however the second I noticed Google Antigravity dive into my Raspberry Pi and code by itself, making fast work of software program issues that had vexed me for months, I used to be hooked. 

Trying again, Google Antigravity was a pivotal “I do know kung fu!” second in my AI journey, and it additionally marked the primary time I hit a utilization restrict. At first, I believed the arcane error code within the Antigravity chat sidebar meant I’d smacked into some form of server error. However like a toddler getting their first ouchie on a scorching range, I quickly realized what actually occurred: I’d blown by means of a multi-hour Gemini utilization quota in a matter of minutes.

Quick-forward to right this moment, as I (and sure loads of different AI customers) hopscotch between Antigravity, Claude Code, and OpenAI’s codex, juggling $20-a-month plans for every service and squeezing as a lot agentic AI coding in as I can with out hitting these damnable limits.

In fact, the large AI suppliers are responding in form, frantically adjusting and readjusting the utilization limits on their flat-rate plans–typically invisibly–as hoards of on a regular basis AI customers unleash armies of AI brokers on their desktops.

Anthropic has most likely drawn essentially the most ire for tweaking its flat-rate plans, quietly dialing down the utilization allowances for Professional ($20 a month) and Max ($100 and up) customers to maintain up with the agentic AI calls for of Claude Code and Claude Cowork, whereas step by step transitioning Enterprise subscribers to a usage-based cost mannequin. 

Google and OpenAI have additionally been busy adjusting their respective charge limits as Antigravity and Codex utilization takes off, with OpenAI lately splitting its Professional tier into $100-a-month and $200/month plans, thus making a center floor for $20/month Plus customers trying to stage up.

Most lately, GitHub has taken the acute course of halting all new signups for its Copilot Professional and Professional+ plans ($10 a month and $39/month, respectively). GitHub additionally introduced that beginning now, it’s going to tighten utilization limits for particular person customers whereas yanking entry to Opus 4.7, essentially the most {powerful} Claude mannequin (save for the “too-powerful-for-you” Claude Mythos) from the Professional tier. 

GitHub, which mentioned it’s going to supply full refunds to subscribers who wish to cancel as a result of sudden modifications, didn’t mince phrases when explaining its new insurance policies:

Agentic workflows have essentially modified Copilot’s compute calls for. Lengthy-running, parallelized periods now usually eat way more assets than the unique plan construction was constructed to help. As Copilot’s agentic capabilities have expanded quickly, brokers are doing extra work, and extra prospects are hitting utilization limits designed to take care of service reliability. With out additional motion, service high quality degrades for everybody.

The true downside, after all, is that AI suppliers from Anthropic and Google to OpenAI and GitHub have been unleashing an increasing number of {powerful} agentic AI instruments on flat-rate customers with (apparently) little regard to whether or not they even have the capability to help them. That results in the large AI gamers regularly messing with utilization limits whereas flat-rate subscribers wrestle to maintain up. 

It’s a nasty expertise for everybody, and I don’t see it bettering anytime quickly.