Goodbye, Cursor. I'm officially getting Codex-pilled.
I know I'm a bit behind the curve here. After seeing the initial hype, I resisted the change, mainly because I have huge production workflows, and I was super comfortable with Cursor. But finally, the migration is happening.
I started using Cursor two years ago, one of its earliest users before it truly hit its PMF traction, and I still love it. To be honest, if cost wasn't an issue, I would still keep using it without a second thought. I'm so entrenched in its workflows that I consider myself a Cursor pro user. I utilize and leverage all its features like workflows, cursor rules, agents, and commands.
The pure reason I'm switching is cost. Empirically, in the last few months, the value I'm getting for my $200 subscription has been consistently reducing. This month, we hit $500 ($200 + API based pricing).
I tried different things, like using auto mode to reduce pricing and other optimizations, but when I'm building, I want to focus on creating, not excessively managing tool costs. It didn't quite work out.
So, out of pure necessity, I tried Codex.
To my surprise, after throwing a couple of random tasks from different repos at it, Codex just works. It’s minimal, hackable, and I’ve already merged PRs with it. That is probably the ultimate test. Also, this bit is hard to quantify and express, but it feels really nice and delightful to use.
I started with the $20 OpenAI Plus plan and was impressed by the value. Today, I'm upgrading to the $200 OpenAI Pro plan. I’m likely riding the VC subsidy wave here too, but as a bootstrapper, you do what you have to do to save money. 🤷
What surprised me the most is how well it's working, even without all my customary agent docs. I'll be migrating all my existing Cursor configurations, like my beloved Cursor rules, to agents.md, which is an open standard and Codex also supports.