Business

Zencoder Lets Developers Bring Their CLI Coding Agent of Choice To Its Platform

AnyFans 2025-09-20

By Frederic Lardinois,Loraine Lawson

Zencoder has long branded itself as an “AI coding agent” for professional developers, but that’s not an easy market to compete in, with even the model builders themselves now launching competing products. So today, Zencoder is switching its model up a bit. The company is still offering its own coding agent, but it will now also allow developers to bring the CLI agents they already pay for (starting with OpenAI’s Codex, Anthropic’s Claude Code and Google’s Gemini CLI) to its platform through its IDE integrations with VS Code and JetBrains.

This, as Zencoder CEO Andrew Filev told me earlier this week, will allow the team to focus on other areas in the development stack where his team can have the most impact, while developers get of using their IDE of choice with all of their built-in debugging and refactoring tools, as well as popular existing features like IntelliSense in Visual Studio Code.

“For two years, we’ve been working on our beautiful UI […],” Filev explained, “Previously, that all worked on our infrastructure. And now, we’re giving the users the choice to still use our UI, but underneath it to use products like Claude or Gemini.”

Every one of these CLI tools, Filev argued, has its own niche, and it helps that Gemini CLI has a free plan, for example, to give users a chance to play with it without committing. Meanwhile, he also noted that OpenAI’s Codex continues to improve, both as a model and as an agentic harness, especially now with the launch of GPT-5-Codex. And Anthropic, at least for the time being, still has the best overall coding model in Opus, but it charges accordingly.

Filev argues that many developers now juggle multiple agents in the command line, with fragmented rules and Markdown files to keep them on track, all held together by less-than-ideal IDE integrations. But with these third-party agents now handling the coding tasks, the Zencoder team gets to focus on other areas.

Filev stressed that Zencoder wants to focus on some of the other areas in the development process, especially its agents for testing, reviews and enforcing consistency across the codebase. Zencoder also allows developers to build custom agents and share them across organizations and in the future, developers will also be able to take those agents — say, you built a great code review agent that is better than Zencoder’s own — and deploy them into the CI infrastructure.

Another area the Zencoder team has focused on is multi-repository intelligence, helping the AI tools understand the entirety of the codebase, including for large monorepos, which some other agentic coding tools often struggle with.

Zencoder also offers additional analytics features to help organizations understand how their developers are using these AI tools and measure their ROI.

Since the team built the IDE integrations on a JSON-RPC architecture, Zencoder will be able to support new CLI tools as they emerge.

Why Change?

There is also an economic aspect to this move as well, of course. The standard $20 starter plans that most services now offer have the worst returns, Filev said. “They’re awful because when you start selling to enterprise, first of all, your price point is higher but your utilization is actually lower,” he said. “When you’re selling something like Cursor to consumers, they count those dollars, so they’re going to squeeze everything from you up until the last limit. And once they’re done, they’re going to start their Windsurf trial and their Zencoder trial and their free Gemini.”

Because of this, companies like Zencoder have to spend a lot of time chasing users who sign up for dozens of free trials, using different email addresses and VPNs and other means. “It’s almost like PayPal in the early days, where you start chasing the fraudsters,” Filev said.

And while Filev admitted that making this change in how Zencoder operates was painful, he believes it’s necessary. “The way I run the businesses, periodically I try to rethink everything,” he said. “I try to say: Okay, if I build the business today, what would be the best business to build? And second, I’ve been building the business for the last two years, so that gives me an incredible foundation. I’ve got the right team, I’ve got the right code. I’ve got the right audience and I’ve got the right brand power. Can I use that as a multiplier for what I’m doing going forward?”

Cursor and others left a vacuum in the enterprise, he argues, because the company didn’t go over that business soon enough and many of the existing tools don’t allow enterprises to run them inside of their own security perimeters, something Zencoder is now focusing on. With Gemini coming to on-prem with Google Distributed Cloud and open models becoming increasingly competitive, enterprises now have a choice to run these models inside their own clouds. This new architecture, which the company is launching today, allows Zencoder to support these deployment methods as well.

Zencoder’s new platform is available today and if you already have a Codex, Gemini or Claude Code subscription, you can get started with just a few clicks.