Swiss Startup Logicstar depends on participating in AI agent games. The startup set in the summer of 2024 is not a more typical AI agent use case in the co -development of code, but to provide a tool to the developer market that can provide autonomous maintenance of software applications. , I got a 3 million dollar -up funding.
Logicstar's CEO and co -founder BORIS PASKALEV (photo in the upper right, functional image, photograph with a colleague co -founder) may have a Startup AI agent in partnership with code development agents such as Cognition Labs. It suggests that there is. Devin -Win -win in business.
Code Fidelity, like human developers, is a problem for the construction and deployment software of AI agents, and Logicstar automatically picks up and fixes bugs anywhere in the deployed code. I want to paint with grease.
At present, Pascalev suggests that even the best models and agents suggest that most of the presented bugs cannot be solved. Dream of boring app maintenance.
For this purpose, they take a model on a large -scale language model (LLMS), such as Openai's GPT and China Deepseek, taking an approach that does not depend on the platform. This allows Logicstar to soak in different LLMs and maximize the Utility of the AI agent.
Paskalev claims that the established team has a technical and domain -specific knowledge, and has built a platform that can solve the problem of programming that can be challenged or in trouble. Masu. They have also succeeded in the past entrepreneurs: in September 2020, launching his previous code review and deeply coded to Cyber Security Giant SNYK.
“Initially, we were thinking of actually building a large -scale language model in the code,” he told Techniccrunch. “Then we realize that it will be a product immediately … Now we are building all of these large language models there, assuming that there are. Assume that there is [AI] Code agent, how do you extract the maximum business value from them? “
He stated that this idea was built based on the team's understanding of how to analyze the software application. “If you combine it with a large language model, focus on the large language model and the AI agent actually contact and verify.”
Test -drive development
What really does it mean? According to Paskalev, Logicstar uses the “Classic Computer Science Method” to analyze the applications that develop the technology to build “knowledge -based”. This allows AI agents to get a comprehensive map of software input and output. How to link a variable to the function. Other links, dependencies, etc.
Next, the AI agent can determine which part of the application will affect all the presented bugs. This allows Logicstar to narrow down the functions that need to be simulated to test the potential correction score.
With this “minimized execution environment” for each Pascalef, AI agents can perform “thousands of” tests aimed at reproducing bugs and identifying the “disability test”. stick.
He has confirmed that the actual bug fix is supplied from LLMS. However, LogicStar platforms have a large -scale work, separate wheat from chaffs, and provide LLMS with the best shortcuts that LLMS can provide in order to enable this “very fast executive environment”. Can be provided.
“What we see [LLMs are] Ideal for prototyping, testing, etc., but it is absolutely good [code] Production and commercial application. I think we are far away from it. This is what our platform provides, “he claimed. “In order to be able to extract these functions of the model today, we can safely extract commercial value and save time for developers to concentrate on important things.”
Enterprise is set to be the first target of Logicstar. The “Silicon Agent” is a part of the salary required to hire human developers, handle the maintenance tasks of various apps, and release more creative and/or challenging engineering talents. However, the purpose is to work with the corporate development team. work. (Or, well, at least until LLMS and AI agents get more abilities.)
The startup pitch advertises the maintenance function of the “completely autonomous” app, but Paskalev makes a review (and supervision) of human developers calling the AI agent by the platform. Check that. Therefore, trust can be earned first.
“The accuracy distributed by human developers is in the range of 80-90 %. Our goal [for our AI agents] He's exactly what he's there. “
It is still an early era for Logicstar. The technology alpha version tests many private companies that Paskalev calls the “design partner.” Currently, Tech supports only Python, but the expansion to TypeScript, JavaScript, and Java has been claimed to be “coming soon”.
“Main goals [with the pre-seed funding] Paskalev adds a technology operation with a design partner that focuses on Python. ” “We are already spending a year, and we have many opportunities to actually expand, and that's why we are trying to concentrate it first.
The prior to the startups before the startup was led by European VC companies Northzone, and Angel investors of Deepmind, Fleet, Sequoia Scouts, SNYK, and Spotify also participated in the round.
In the statement, Michiel Kotting, a partner of NORTHZONE, states: “AI -led code generation is still in the early stages, but the productivity I have already seen is innovative. This technology streamlts, reduces costs, and accelerates innovation. Logicstar has been reconstructed to have a huge amount of technical knowledge and achievements of the team. It plays an important role in software maintenance.
Logicstar runs a waiting list for potential customers who want to be interested in acquiring early access. It turns out that the beta release is planned in the latter half of this year.