Today I have some news to share on the future of Plandex.
Cutting to the chase: I've recently accepted an engineering position at Promptfoo, and I've made the tough decision to begin winding down Plandex as a company and cloud service.
The open source project will remain available on GitHub.
Here are the details:
- New trials and sign ups on Plandex Cloud have been disabled
- All subscriptions have been paused and will no longer renew
- Credit purchases and auto-rebuys have been disabled
- Plandex Cloud will continue operating for existing customers until November 7th so that you have time to use up any accumulated credits
- On November 7th 2025, Plandex Cloud will shut down
- The Plandex open source project will remain on GitHub
Now I'll talk a bit about Plandex's history and how I came to this decision.
When I first started working on Plandex in late 2023, GPT-4 API access had only been widely available for a few months, and the context limit was 8k tokens. In spite of the limitations, I was fascinated by the possibilities.
Having already used ChatGPT, via the browser, to do projects that would have been impossible or impractical prior to AI, I imagined a much better workflow than copying and pasting code back and forth between my IDE and the browser. I already knew that with the right prompting, ChatGPT could accomplish some impressive tasks, even if they were generally limited to a single file and a moderate level of complexity. I couldn't help but wondering what it could accomplish if given the ability to edit files directly and chain multiple responses.
Plandex wasn't the first take on this idea. Aider was already quite popular, and there were a number of other interesting tools in the space as well.
I decided to focus on a few areas that I hoped would differentiate Plandex from the pack: deeply embedded version control, an agentic loop, and a focus on large, multi-file tasks.
The initial v1 launch was promising: Plandex was the top repo of the day (and week) on GitHub. The repo got thousands of stars and thousands of people tried the cloud version. Hundreds of people joined Plandex's discord. The enthusiasm was palpable.
As an engineer, I was having a blast working on the project and trying to push the limits on what AI codegen could accomplish. But having a family to feed and being, rather unfortunately, not independently wealthy, I knew I also needed to build a working business to make it sustainable.
Having been a solo founder of a previous startup, which I worked on for years, I had no illusions about how difficult this would be, especially in a space which was shaping up to be highly competitive, with many large and well-funded players. But based on the v1's success, I was able to raise some pre-seed capital from a terrific VC fund, and I resolved to give it my all.
Fast forward to today, and Plandex has come a very long way. Plandex v2 launched, making it one of the first coding agents capable of full autonomy.
That said, the competition has only gotten more fierce. Apart from the pre-existing players, the foundation labs started jumping into the mix with their own dedicated coding agents (most notably, Anthropic's Claude Code and OpenAI's Codex).
While there were definite bright spots for Plandex, including two months after the v2 launch that saw Plandex Cloud 10x-ing and then tripling in revenue, I started to feel that I was stretched impossibly thin trying to keep up with the cutting edge on both engineering and marketing. Churn increased quickly as Claude Code became the dominant tool in the space and began to heavily subsidize tokens with their Pro and Max plans.
I ultimately decided, trying to assess the situation honestly, that Claude Code had executed masterfully on the original vision of Plandex, and that it was pointless trying to play catch up as a solo founder without employees.
I worked on a pivot for a couple months, building out the alpha of an AI-based project planning web app that could interoperate with Claude Code and other agents for implementation. While I still see quite a lot of potential here, I also did some soul searching and began questioning whether I was really up for another long solo grind.
I realized that the parts of doing a startup I really love are engineering, design, and product. I'm not particularly motivated by growing a business, except as an enabler to spend my time on those things. I also greatly prefer working with others to working solo, but after nearly a decade-long founder career, I haven't been able to achieve the escape velocity needed to put together a stable team.
The conclusion, when I looked at things in those terms, seemed rather obvious: I should join a great team that already has traction, so that I can focus on what I'm best at and love doing. So when I got the opportunity to join Promptfoo, which has put together an awesome group of smart people and whose focus on AI and security neatly aligns with my past experience, I decided to take it.
I want to sincerely thank you for being part of Plandex's journey. I wish you the best in all your AI coding adventures.
Onward,
Dane