---
title: "Running Out of AI"
description: "A developer's journey through hitting AI code generation limits \u0026 exploring alternatives like API billing, other vendors, \u0026 local models."
categories: ["AI","software development"]
keywords: ["AI","Claude","GPT-OSS","coding","API limits","software development"]
ai_summary: "A developer hits their Claude code credit limit \u0026 explores various alternatives, weighing the costs \u0026 benefits of switching to API billing, trying different AI models like Gemini \u0026 Codex, or setting up a local GPT-OSS model."
date: 2025-11-12
lastmod: 2026-07-28
canonical_url: https://www.tomtunguz.com/running-out-of-ai/
author: "Tomasz Tunguz"
---


By Monday lunch, I had burned through my Claude code credits. I'd been warned ; damn the budget, full prompting ahead.

{{< email_image src="sgoagcamztooxmkomw2a" alt="Screenshot 2025-11-12 at 8.25.37 AM" width="540" height="432" >}}
I typed ultrathink to solve a particularly challenging coding problem, knowing the rainbow colors of the word was playing with digital fire.

{{< email_image src="vjvqgkqduzerdsbvlp6u" alt="Screenshot 2025-11-12 at 8.26.41 AM" width="540" height="243" >}}

When that still couldn't solve the issue, I summoned Opus, the biggest & most expensive model, to solve it.

Now two days on, I've needed to figure out alternatives. Do I :

- Switch to API billing (how much will that cost?)
- Try another vendor? Gemini's model is great, but ageing ; at nearly 8 months old, its a capable jalopy. Cursor's free coding model Composer 1 sprints at problems with aplomb but a bit overwhelmed at times. Codex, the plodding giant is brilliant at large-scale technical challenges.
- Create another Max subscription & switch between them? Can I ask AI to write a script to save me the hassle of changing my identity?
- Stand-up GPT-OSS to run locally? A little bit more latency but potent & twice as fast on llama.cpp compared to Ollama.
- Go back to writing code the old way? The hedonic treadmill moves quickly. I tried to return to the old ways, but it was painful. I've already forgotten where the blog server script is. Claude? Do you remember? Claude?

I'm working through the math of which option will cost more. How much is the Max plan subsidized? Will knowing the true API cost of my Claude Code usage increase my willingness to pay?

Switching between tools incurs costs. The tools, the workflow, the prompts that I've optimized for Claude code must all be ported (at my expense!) to other tools.

As the capabilities of these models begin to plateau, the costs to shift increase. So does my willingness to pay for Claude to answer me.