CodiumAI eyes ‘agentic future’ for AI-powered software developers
Every company is a data company.
Now, of course, every company is an Artificial Intelligence (AI) company.
Plus, logically every company must now think of itself as a generative AI company.
But how many generative AI company contenders also produce platform-level tools for software application development professionals to create AI-driven, AI-enriched, AI-augmented or AI-accelerated (other adjectives are also available) code that runs at high-quality levels across a significant user base?
This (surely arguably) smaller definition of how AI know-how can be applied to any given company may be applicable to Tel Aviv-based CodiumAI, a generative AI company focused on the developer codebase. The firm has now announced today the deployment of a proactive AI Agent to all 500K users of its software development tools.
The semi-autonomous solution works in tandem with popular code completion tools to enhance the quality of AI-generated code, by proactively suggesting improvements to developers as they code. Alongside the agent, CodiumAI is also releasing its own code completion tool that emphasizes code integrity.
According to CodiumAI, Code generation, powered by generative AI ‘is ubiquitous’ with 92% of developers already using AI tools to assist in writing code.
Developers code – and…
But, as we know, there is more to a developer’s role than just writing code. Creating accurate documentation, improving existing code, and exhaustive testing are cornerstones of the profession and are overlooked by current AI assistant technology.
While AI assistants are helping developers to write code quicker, there are nagging concerns that they could make software quality issues worse.
The Codiumate Agent runs alongside developers as they work, suggesting relevant tests, drafting documentation and identifying code duplicates without prompting. By offering assistance to developers that complements completion tools, the AI Agent allows code faster and (in theory) better.
Our agentic future
Designed to enhance CodiumAI’s existing code integrity and testing offering, the agent brings testing and validation directly into their existing development environment.
“While coding assistants like GitHub Copilot are extremely popular, more and more developers are insisting on capabilities beyond what is available,” said CodiumAI CEO and co-founder Itamar Friedman. “Developers want AI to help them with the often tedious tasks that are critical to ensuring code quality and correctness and these go beyond just writing code. We at CodiumAI are looking to an agentic future [i.e. one where more software agents are used) where busy developers will team up with agents to improve coding quality and processes. Integrated agents will be there every step of the way, offering insights, recommendations and solutions proactively.”
This announcement builds upon the recent success of CodiumAI’s AlphaCodium research, which proposed a new approach to code generation by LLMs, one built upon a test-based, multi-stage iterative flow to solve code problems. AlphaCodium’s performance on the CodeContests benchmark showed that its performance improved GPT-4’s accuracy from 19% to 44%.
Developers can try out the coding agent by installing Codiumate for VS Code or JetBrains IDEs, start with a Google search of CodiumAI.
CodiumAI tweets at @CodiumAI
