LogicStar, a Swiss startup founded in the summer of 2024, is making strides in the realm of AI agents. The company has secured $3 million in pre-seed funding to develop tools that offer autonomous software maintenance for developers, shifting focus from the usual AI agent role of co-developing code.
According to LogicStar’s CEO and co-founder, Boris Paskalev (featured in the image at the top right with his co-founders), the AI agents created by the startup may one day collaborate with code development agents—like Cognition AI’s Devin—creating a mutually beneficial business relationship.
Code fidelity remains a challenge for AI agents that are responsible for software creation and deployment, much like it is for human developers. LogicStar aims to facilitate the development process by identifying and fixing bugs in deployed code automatically.
Currently, Paskalev points out that “even the top models and agents” struggle to address most bugs they encounter, presenting an opportunity for a startup focused on enhancing these capabilities and fulfilling the desire for more efficient app maintenance.
Their approach builds on large language models (LLMs) such as OpenAI’s GPT and China’s DeepSeek, taking a model-agnostic perspective for their platform. This strategy allows LogicStar to leverage various LLMs and optimize the utility of its AI agents based on which foundational model best resolves specific coding challenges.
Paskalev asserts that the founding team possesses the technical expertise and industry knowledge required to create a platform capable of tackling programming issues that often stump LLMs working independently. Their previous entrepreneurial success adds credibility to their venture: Paskalev sold his former code review startup, DeepCode, to cybersecurity leader Snyk in September 2020.
“Initially, we considered developing a large language model specifically for code,” he shared with TechCrunch. “However, we realized that this would soon become commonplace… Now, we’re constructing our platform with the assumption that these large language models will be readily available. Considering that there are indeed efficient [AI] agents for code, how can we extract maximum business value from them?”
This concept builds upon the team’s deep understanding of software application analysis. “When combined with large language models, the focus shifts to verifying and grounding the suggestions made by those models and AI agents,” he explained.
Test-Driven Development
In practical terms, Paskalev explains that LogicStar analyzes each deployed application using “classical computer science methods” to create a “knowledge base.” This knowledge base equips the AI agent with a thorough understanding of the software’s inputs and outputs, connections between variables and functions, along with other dependencies and links.
For every bug it encounters, the AI agent can identify the affected sections of the application, allowing LogicStar to focus on testing numerous potential fixes by simulating necessary functions.
According to Paskalev, this “minimized execution environment” empowers the AI agent to conduct “thousands” of tests to reproduce bugs and pinpoint a “failing test.” This “test-driven development” methodology ultimately leads to effective solutions.
While actual bug resolutions are derived from LLMs, LogicStar’s platform facilitates a “very fast execution environment,” enabling its AI agents to work efficiently at scale and present users with the best possible options from LLMs.
“What we find is that LLMs are excellent for prototyping and testing, but they fall short for [eventual] production and commercial applications. I believe we still have a long way to go, and this is precisely what our platform aims to provide,” he argued. “By leveraging today’s model capabilities, we can safely extract commercial value and allow developers to concentrate on more critical tasks.”
LogicStar’s primary targets are enterprises. Its “silicon agents” are designed to collaborate with corporate development teams, performing various application maintenance tasks at a significantly lower cost than hiring a full-time human developer, thus allowing engineers to focus on more innovative or intricate projects—at least until LLMs and AI agents evolve further in capability.
Though the startup claims its technology can achieve “fully autonomous” app maintenance, Paskalev reassures that the platform will enable human developers to review (and supervise) the fixes proposed by its AI agents, ensuring that trust is built over time.
“The accuracy provided by a human developer ranges from 80% to 90%. Our target for our AI agents is to reach that same level,” he adds.
LogicStar is still in its infancy: an alpha version of its technology is being piloted with several undisclosed companies, referred to as “design partners.” Currently, the platform only supports Python, with plans to extend support to TypeScript, JavaScript, and Java soon.
“The primary objective for the pre-seed funding is to demonstrate that the technology works effectively with our design partners—initially focusing on Python,” Paskalev clarifies. “After a year of development, we see ample opportunities for expansion, and we’re keen to prove its value in this specific instance first.”
The startup’s pre-seed funding round was led by the European venture capital firm Northzone, with contributions from angel investors affiliated with DeepMind, Fleet, Sequoia scouts, Snyk, and Spotify.
In an official statement, Michiel Kotting, a partner at Northzone, remarked: “AI-driven code generation is still in its infancy, yet the productivity enhancements we’re witnessing are remarkable. The potential for this technology to refine development processes, diminish costs, and hasten innovation is substantial. The team’s extensive technical prowess and proven track record position them to yield genuine, meaningful outcomes. The future of software development is evolving, and LogicStar will be a key player in the domain of software maintenance.”
LogicStar is currently managing a waiting list for prospective customers eager for early access and has indicated that a beta release is anticipated later this year.
TechCrunch offers an AI-focused newsletter! Sign up here to receive it in your inbox every Wednesday.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
