LogicStar is building AI agents for app maintenance

LogicStar, a Swiss startup founded in summer 2024, has secured $3 million in pre-seed funding to develop AI agents for autonomous software application maintenance. This focus differentiates them from the more common use case of AI agents for code co-development. CEO and co-founder Boris Paskalev envisions their AI agents potentially partnering with code development agents, such as Cognition AI’s Devin, creating a mutually beneficial ecosystem.

Recognizing that code fidelity is a challenge for AI agents, LogicStar aims to streamline development by automatically identifying and fixing bugs in deployed code. Paskalev notes that current models and agents struggle to resolve most bugs, presenting an opportunity for a dedicated AI bug-fixing solution.

LogicStar’s platform is built on top of large language models (LLMs), taking a model-agnostic approach to maximize agent utility by selecting the most appropriate foundational model for each specific code issue. Paskalev emphasizes the team’s technical expertise and prior success (his previous startup, DeepCode, was acquired by Snyk) as key to building a platform that can effectively address programming challenges that often stump LLMs alone. Instead of building their own LLM, LogicStar aims to leverage existing ones and extract maximum business value.

Their approach involves analyzing each application to create a “knowledge base” using classical computer science methods. This map of inputs, outputs, variables, functions, and dependencies allows the AI agent to pinpoint the impacted areas for any given bug. A “minimized execution environment” then allows the agent to run thousands of tests to reproduce the bug and identify a “failing test.” This “test-driven development” approach helps the agent arrive at a reliable fix.

While the actual bug fixes are generated by the LLMs, LogicStar’s platform provides the fast execution environment necessary to efficiently filter and select the best solutions. Paskalev argues that while LLMs are useful for prototyping, they are not yet reliable enough for production, commercial applications. LogicStar’s platform aims to bridge this gap, extracting commercial value from LLM capabilities and saving developers time.

Initially targeting enterprises, LogicStar plans to deploy its “silicon agents” alongside development teams to handle routine maintenance tasks, freeing human developers for more complex work. While touting “fully autonomous” maintenance, Paskalev confirms that human review will be possible, emphasizing the importance of building trust. Their accuracy goal is to match the 80-90% range of human developers.

Currently, an alpha version is being tested with undisclosed “design partners,” focusing on Python support. Expansions to TypeScript, JavaScript, and Java are planned. The pre-seed funding will be used to prove the technology’s effectiveness with design partners, focusing initially on Python.

The pre-seed round was led by Northzone, with participation from angel investors from DeepMind, Fleet, Sequoia scouts, Snyk, and Spotify. Northzone partner Michiel Kotting highlighted the revolutionary potential of AI-driven code generation and LogicStar’s expertise in reshaping software maintenance.

LogicStar is operating a waiting list for early access and plans a beta release later this year.

Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *