By Shiri Epstein, NoCamels -
Programs, apps and websites are all built using code that tells a computer, phone or other device what to do and how to do it. Actually creating that code is a tricky and complicated process, where even the smallest error can cause entire platforms to fail.
Finding a mistake in rows of letters and numbers can be even harder and is extremely time consuming. In fact, software developers can spend up to 50 percent of their working hours testing and debugging code, according to well-respected trade magazine ACM Queue.
But now an Israeli company has developed an AI “team member” to test and review code to make sure it is all working as it should, rooting out any potentially catastrophic missteps.
Codium co-founder and CEO Itamar Friedman tells NoCamels that the startup decided to take a “bottom up” approach and comb through existing code to find the mistake rather than papering over any cracks – something that has become all too common as programmers work to tight deadlines and under high pressure.
Friedman says that Codium wanted to be connected to the code in order to see where the problem was coming from.
“We’re focused on solving bugs and issues; we’re focused on testing code,” he explains. “While most coding assistants are trying to push you towards writing more lines of code and replacing your lines of code, we are there to empower you, to make sure that your lines of code are actually working as expected.”
Friedman and Codium co-founder and CPO Dedy Kredo believed that using artificial intelligence as a team member that could test and debug code was crucial both to supporting developers and taking programming into the future.
They achieved this with two different AI plugins (downloaded extensions to a program) – each performing a series of different yet vital tasks.
The first of the two is Codiumate, a plugin for an integrated development environment, a kind of digital toolbox containing multiple features to help developers, such as debugging, tutorials and even a spellchecker.
Codiumate can perform multiple simple commands for existing code, such as run a new test of it, enhance it and even rewrite it to remove any bugs.
It can also perform more advanced tasks, such as scanning code without running it, in order to tell developers how it will behave. And according to Friedman, this provides the developer with a simple, easily understood explanation of how the software performs.
“You get a natural language description of what your software should do [that is] easier to read – even for the developer, and then you can generate a test for that,” he says.
The second of the plugins deals with pull requests – when a developer is ready to integrate new code into an already existing system and asks for it to be checked for any bugs first.
Codium’s open source PR-Agent plugin is designed to assist with pull requests – walking through the code, testing it and summarizing it in order to outline possible issues that might arise.
Based on the needs of each client, the plugin can include more extensive testing to create recommendations of how to improve the code and root out potential bugs.
“We don’t say 100 percent code coverage as our vision. We say zero bugs,” explains Friedman.
All of the suggestions made by the AI are left to the developers to action. Nothing, Friedman stresses, is automatically changed or rewritten by the platform.
A senior official from OpenAI was among the investors in the company, which was founded in 2022. Other funding came from Israeli venture capitalist company TLV Partners, New York-based Vine Ventures VC and global fund MyVentures, all of which invest in early-stage startups.
The company announced in early 2023 that it had raised $11 million in seed funding.
Less than two years after its establishment, Codium’s plugins have already been installed by over half a million developers around the world and more than 1,000 teams are already using the licenced program that allows them to collaborate in real time.
The Tel Aviv-based company recently won praise for its research paper outlining how AI can be integrated into the process of code generation. This inclusive approach is called AlphaCodium, in honor of the open-source AI coding tool called AlphaCode.
“AlphaCodium is taking the developer’s best practices on how to develop software and engineering it into AI being able to reproduce that,” Friedman says.
Among those praising the paper was Andrej Karpathy, one of the co-founders of OpenAI and the former director of AI at Tesla.
For Friedman the future is to keep creating tools that would make the life of developers easier – and that means further integration of artificial intelligence.
“AI is not only going to be code completion, instead AI can empower the software development landscape,” he says.
“That’s going to happen and cover the entire software development lifecycle.”
