Codium announces Codiumate, a new AI agent that seeks to be Devin for enterprise software development

Join us in Atlanta on April 10th and explore the landscape of security workforce. We will explore the vision, benefits, and use cases of AI for security teams. Request an invite here.


One of the most well-read VentureBeat articles in recent weeks has been our coverage of Devin, an AI “software engineer” from startup Cognition that can autonomously write code, even full applications, given natural language instructions from a user.

It makes sense, as this type of capability would be hugely helpful for software developers, managers, and those who employ them. Already, leading AI influencers such as University of Pennsylvania Wharton School of Business professor Ethan Mollick have posted examples of Devin effectively coding together entire websites for them.

But for enterprises looking to maintain existing codebases and apps, and push updates to their software for employees or customers in high-security, highly regulated industries such as finance, healthcare, law, government, and telecommunications — the idea of turning loose an AI software engineer all on its own to do these tasks may be slightly less appealing.

Another new Israeli startup, Codium AI, is trying to make it possible for enterprises to reap the benefits of autonomous AI coding while also still retaining human oversight and compliance with regulations and existing software.

VB Event

The AI Impact Tour – Atlanta

Continuing our tour, we’re headed to Atlanta for the AI Impact Tour stop on April 10th. This exclusive, invite-only event, in partnership with Microsoft, will feature discussions on how generative AI is transforming the security workforce. Space is limited, so request an invite today.

Request an invite

Today, the company announced “Codiumate,” a new AI agent described as “semi-autonomous.” It works alongside a human software developer or team. It takes any existing code snippets a human developer highlights in their environment and automatically drafting step-by-step development plans based on it, writes code according to the plan, identifies duplicate code, drafts documentation for the code so that developers can reference it going forward, and suggests tests to ensure the code works properly before it is deployed for real in a live, production environment.

In fact, writing code is just “20 to 30 percent of what developers do,” said Codium co-founder and CEO Itamar Friedman in a video call interview with VentureBeat, citing surveys such as one from The New Stack and Tidelift on how engineers spend their time. “All the rest is meant to define what we do with our code, what we want to do with it, and check that it actually happened. That’s what we’re focused on at Codium AI.”

How Codiumate works alongside and with developers

[embedded content]

Friedman demoed the tool for me live during our call, showing how the user can drag their mouse cursor to highlight multiple code blocks in their development environment and then instruct Codiumate to generate a plan for how they want to use that code.

The agent presents the plan to the human user for review, which the human developer can then edit and adjust at will — any portion of it.

Once a user is happy with the plan, the human developer user can click a button labeled “start implementing the plan,” and the Codiumate AI Agent will begin coding it for them, though it will occasionally stop to ask the human user if what it is doing is correct, and allow time for the human user to edit the code or fix any issues as they see fit. It may also prompt the human developer to complete code or write code on their side in order for it to continue. You can see a prerecorded demo uploaded by Codium to YouTube embedded above.

In this way, Codium hopes that Codiumate will aid developers in their workflow, speeding up all the manual typing they would otherwise have to do, doing the “heavy lifting” and mechanical coding work, while enabling the developer to act more as a hands-on product manager overseeing the process and course correcting it as necessary, almost as though it is a junior developer or new hire to the team.

The technology powering the Codiumate agent on the backend is “best of breed” OpenAI models, according to Friedman. The company is also experimenting with Anthropic’s Claude and Google’s Gemini. It also offers its own large language model (LLM) designed with its AlphaCodium process that increases the performance of other LLMs in code completion tasks.

While the former is available to all users, the latter Codium LLM is only for paying enterprise users. Friedman said it is superior to OpenAI’s models on coding and that a “Fortune 10” company that could not be named due to confidentiality reasons was already using it in production.

Zero data retention

Addressing a growing market of AI tools that seek to offer security and anonymity to enterprise customers, Codiumate also practices “zero retention,” meaning that any code or information entered into it will be deleted immediately after the human user closes or ends their session with the tool.

The company also complies with industry best-practices and standards such as SOC 2. It can run Codium on a company’s private servers or even on air-gapped computers for those looking to have a completely on-device experience.

Pricing and customers

Codiumate is being offered as part of a growing library of AI development agents from Codium AI — previously, the company released a PR Agent for automating pull request analysis and feedback.

The company offers a free tier powered by OpenAI’s GPT-3.5, a Teams tier that costs $19 per user per month, and an Enterprise tier with variable software-as-a-service (SaaS) subscription pricing depending on the customer’s unique implementation and needs.

Already, it claims 500,000 users globally, according to a press release from the company.