Fellow annoyed Google AI Pro subscriber here!

Can confirm, I initially enjoyed the 5-hour limits on Gemini CLI and Antigravity so much that I paid for a full year, thinking it was a great decision

In the following months, they significantly cut the 5-hour limits (not sure if it even exists anymore), introduced the unrealistically bad weekly limit that I can fully consume in 1-2 hour, introduced the monthly AI credits system, and added ads to upgrade to Ultra everywhere

At the very least the Gemini mobile app / web app is still kinda useful for project planning and day-to-day use I guess. They also bumped the storage from 2TB to 5TB, but I don't even use that

It should be illegal to change the terms of the subscription mid-period. If you paid for the full year, you should get that plan for the whole year. I don't understand how it's ok for corporations to just change the terms mid-way, and we just have to accept it.

> It should be illegal to change the terms of the subscription mid-period

Unfortunately, at least for those of us in the US, there isn't legally much that can be done. It's simply not possible to make a contract that would obligate a company to fulfill its promises on this type of sale.

I'm sure the T&C say something like "you're going to pay us money, and we reserve the right to give you something for it, or maybe nothing, and you should thank us for the privilege".

Don't bother upgrading to ultra. It's also now easy to burn all your credits where in Jan it was almost impossible

It's the exact same thing they did with Google BigQuery, which initially was an absolutely amazing piece of technology before they smothered it with more and more limits and restrictions. It's like they're putting SREs first, customers second.

My personal experience is way different: I struggle to burn through more than 50% of the 5 hour limit

For context, with Google AI Pro, I can burn through the Antigravity weekly limit in 1-2 hours if I force it to use Gemini 3.1 Pro. Meanwhile Gemini 3 Flash is basically unlimited but frequently produces buggy code or fail to implement things how I personally would (felt like it doesn't "think" like a software dev)

I also tried VS Code + Cline + OpenRouter + MiniMax M2.7. It's quite cheap and seems to be better than Gemini 3 Flash, but it gets really pricy as the context fills up because prompt caching is not supported for MiniMax on OpenRouter. The result itself usually needs 3-6 revisions on average so the context fills up pretty often

Eventually I got Claude Max 5x to try for a month. VS Code + Claude Code extension on a ~15k lines codebase, model set to "Default", and effort set to "Max". So far it's been really good: 0-2 revisions on average, and most of the time it implements things exactly how I would or better. And, like I said, I can only consume 40-60% of the 5-hour limits no matter how hard I try

Granted, I'm not forcing it to use Opus like OP (nor do I use complicated skills or launch multiple tasks at the same time), but I feel like they really nailed the right balance of when to use which model and how to pass context between the them. Or at least enough that I haven't felt the need to force it to use Opus all the time

Reading the other negative comments makes me wonder if this is only because I'm getting a hidden newcomer's limit bonus or something though hahah

it may also be local/timezone effects

it has been reported that it behaves very differently depending on those factors, presumably because people are placed in best-effort buckets, who knows

Same here: I found the multibuffers feature really useful, but the extension system really couldn't hold a candle to VS Code at the time of my testing

Spent a couple of hours trying to make the Svelte extension ignore a particular type of false positive CSS error, failed, and returned to VS Code

Will definitely give it another chance when the extension system is more mature though!

I think one of the less mentioned benefit of coding agents these days is how much easier it is to do big migrations like these

Recently I was ~70% done on a project using the relatively young Electrobun framework when I hit a non-negotiable limitation

So I told a $$$ agent to plan and implement a migration to Tauri, then repeated the loop of telling a $ agent to look for feature parity issues and having a $$ agent verify and fix the issues

In a couple of hours I got virtually the same app in a different framework

So there's definitely less burden in choosing the right framework at the start of a project, and less justification to keep a suboptimal infrastructure simply due to cost of migration

Seems interesting, but I wonder how this would be better than just asking an LLM to implement the skeletons?

For most components, current generation models should be able to understand the component code and produce skeletons that occupies exactly the same space

I do this often, and they do! Yours is the same question I had.

I'd say it might be due to external factors such as a bad working environment

Someone who originally has coding as their passion, for example, might eventually come to dislike it due to overwork. And in doing so they overcompensate by imagining that the total opposite of office work, e.g. farming, would be a better way of life, even though it may not necessarily be true

That said, I think something like a week long course of farming targeted towards white collar workers, with all of the "fun and refreshing" parts but only educational exposure to the painful parts would be a great business idea (or maybe something like that already exists somewhere)