I’ve been a paying Claude Code customer since it launched in May 2025, which makes me about as long-standing a customer as you can get. It set the bar for agentic coding CLIs. Everything that came after has been measured against it. And for most of that time, nothing else came close.

To be clear, none of this is a knock on Claude Code. It’s about why I’ve had to start looking for an alternative, despite how good Claude Code is.

The usage caps are the core issue. If you use AI occasionally, you’ll never hit them. But if you’ve committed to an agentic, AI-first way of working, where Claude Code is running close to full-time and handling real chunks of your development workflow, you will hit them. Regularly. The only escape is API billing, which at that level of usage is prohibitive for many developers. The problem isn’t that heavy usage costs more. That’s fair. The problem is that Anthropic’s pricing simply doesn’t work at the scale that serious AI-first development demands.

That led me to start testing alternatives. I wrote yesterday about how the non-Anthropic models compare when you actually use them for real work. The short version: Codex is usable, Minimax-2.5 and GLM-5 are not there yet, and Opus and Sonnet are still clearly ahead. I did get some of those models running inside Claude Code CLI via providers that expose an Anthropic-compatible API, but it’s not something Claude Code is designed for. It’s built around Anthropic’s models, and that’s fair. It just meant I needed to find a harness that supports model flexibility by design, not as a hack.

From there I looked at other harnesses. OpenCode, Kilo Code, Gemini CLI. None of them felt viable. Tool use across longer sessions and end-to-end features just isn’t where it needs to be with any of them, and using them day to day would have been a downgrade.

I stumbled across Factory AI’s cli tool after testing everything else on that list. So far, I have not found that I’m leaving anything on the table by using it instead of Claude Code. The transition was straightforward, the shortcuts are similar, and the general workflow is close enough that there’s very little muscle memory to rewire. In some ways it’s actually a step up.

Where Factory AI shines is the model flexibility. Via Vibe Proxy, I can use my existing Anthropic subscription and my OpenAI subscription from the same harness. I’m not walking away from Opus and Sonnet. I’m just no longer locked into Claude Code to access them.

I’ve fully moved off Claude Code CLI for now. Factory AI in the terminal is my daily driver. It’s early days, and this is still a work in progress. But so far the developer experience feels comparable, and I can use the same skills, rules, and repo setup with any model without changing my workflow. That matters, because it means I can save my Opus and Sonnet usage for the work that actually needs it and fall back to Codex when I’ve burned through my weekly limits. Codex is free right now, and as I wrote yesterday, it is usable.

The model matters. But being locked into one way of accessing it, with usage caps you can’t work around, is a problem that eventually outweighs everything else. Factory AI gave me a way out without making me start over.