VentureBeat readers have taken an enormous interest in our coverage of Devin, an AI “software engineer” from startup Cognition that can autonomously write code–even full applications–when given natural language instructions from its users. One of the most read VentureBeat articles recently has been our coverage of Devin.
At first glance, this seems sensible; such a capability would prove invaluable for software developers, managers and those employing them. Already leading AI influencers like University of Pennsylvania Wharton School of Business professor Ethan Mollick have posted examples of Devin successfully coding entire websites for them.
Companies looking to manage existing codebases and apps while pushing updates out to employees or customers in highly regulated industries such as finance, healthcare, law, government or telecom may find AI software engineers less appealing when it comes to performing these tasks independently.
Codium AI, another new Israeli startup, aims to enable enterprises to reap the benefits of autonomous AI coding while maintaining human oversight and compliance with regulations and existing software.
Today, the company unveiled “Codiumate,” an AI agent it describes as semi-autonomous. This agent works alongside human software developers or teams and takes any existing code snippets highlighted by human developers from their environment to draft step-by-step development plans; writes code according to these plans; detects duplicate code; documents it so developers can refer back to it later; drafts documentation about said code so it can be referenced later; drafts documentation about said code so future developers may refer back; drafts tests before deployment in real world environments;
Codium AI’s co-founder and CEO Itamar Friedman recently stated in an interview with VentureBeat via video call that code writing only comprises about 20-30 percent of their job responsibilities. According to surveys such as those conducted by The New Stack and Tidelift regarding engineers’ time usage. All other activities aim at defining what we intend to accomplish through code, and verifying if our goal has been reached successfully – something Codium AI specializes in.”
Friedman demonstrated the tool live for me during our call, showing how users could highlight multiple code blocks within their development environment before instructing Codiumate to generate a plan on how best to utilize this code.
The agent then presents their plan to the human user for review, who may then change or amend any part of it as desired.
Once users are happy with their plan, human developer users can click a “Start Implementing the Plan” button and the Codiumate AI Agent will start coding for them – stopping every now and then to ask if what it’s doing is correct and allow for editing or fixing of issues as necessary by the human developer user. It may even prompt them to complete code or write new ones on their side if needed in order for Codiumate AI Agent to continue. Below you can watch a prerecorded demo uploaded by Codiumate AI Agent uploaded onto YouTube by Codium above.
