The Cost of AI Is Now My Main Blocker

My main blocker right now is the cost of AI. I use AI for everything I do and I am very productive with it. I am also deliberately working slower than I could, because I cannot afford as much of it as I would need.

All of my coding goes through Codex or Claude Code, for my own side projects as well as for client work. Clients expect me to use AI and to move fast because of it, and I am so used to working this way that there is no going back for me either. Outside of coding, I do my outreach with AI, I do my accounting with AI, and everything on the marketing side.

I pay for two $200 Claude Code subscriptions and two $200 Codex subscriptions, so $800 a month before any API usage. On top of that I spend a few hundred dollars a month on OpenRouter for open-weight models, and I have bills at Modal, Together AI and Fireworks AI as well. It is still not enough.

Until a few weeks ago I had three Codex subscriptions. I cancelled one while I was on vacation, and now I cannot get it back. On September 10, OpenAI paused new sign-ups and upgrades to the $200 Pro plan because of the demand for GPT-6 Astra. I would reactivate that third subscription today if OpenAI let me.

The alternative would be a third Claude Code subscription, and I am tempted. The problem is Fable. On the Max plan, Anthropic lets you use up to 50% of your weekly usage limits on Fable models, and Fable burns through that allowance faster than the other models. Fable is the model I want on the Claude side, so paying $200 for a plan that lets me put only half of the usage into Fable sucks, and it is why I have not bought a third one.

Over the last couple of months I made it a habit to use the cheapest model that can do the job. For automations and simple tasks I use DeepSeek or GPT-5.6 Luna. But that only works for part of my work. For complex codebases, for code reviews, and even for writing, where I care about style, I need the state-of-the-art models.

So I slow down on purpose. I have the workflow, I have a list of things I want to build, and I still sit and wait for a usage reset, or I hand a task to a weaker model and accept a worse result.

For now I keep the four subscriptions, push everything I can to DeepSeek, and wait for OpenAI to reopen the Pro plan so I can get the third Codex subscription back.

What I'm building

Delegate tasks. Get software.

Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.

Take a look at vroni.com