Swiss start Reasonable It is a bending in joining the AI Agent game. In the summer of 2024-established a start has imposed $ 3 million on pre-spades funding to bring tools on the marketplace market that can do autonomous software applications, instead of the most formal AI agent use of co-development code.
Logicstar’s chief executive and co-founder Boris Paskalev suggests that AI agents could end up working with Code Development Agents-such as Cognition Labs Devin Labs-in a business win-win.
Fidelity Code is an issue for building and developing AI software, as well as human manufacturers, and Logicstar wants to do its part to lubricate the growth wheel, automatically and defining errors wherever they can appear in Developed code.
As is the case, Paskalev suggests that “even the best models and agents” out there are unable to resolve the majority of the errors presented – hence the team that spies an opportunity for an AI start -up dedicated to improving them Probability and delivery is the dream of less tiring application maintenance.
To this end, they manufacture over the large linguistic models (LLMS)-such as the Openai GPT or even China’s Deepseek-involving a model-an aggravating approach to their platform. This allows Logicstar to dive into different llms and maximize the AI agents’ utility program, on the basis of which the fundamental model works best to resolve a particular code issue.
Paskalev argues that the founding team has the techniques and specific knowledge of building a platform that can solve programming problems that can challenge or operate outfox llms that work on their own. They also have a past business success to point out: which sold the previous start of the review code, Deepcode, to the cyberspace of Snyk in September 2020.
“At first we were really thinking of building a large linguistic model for code,” he told TechCrunch. “Then we realized that this would quickly become a merchandise … Now we build by assuming that all these great linguistic models are there. Assuming there are some really decent [AI] Agents for the code, how do we extract maximum business value from them? ”
He said that the idea based on the team’s understanding of how to analyze software applications. “Combine this with large linguistic models – then focus on ground and verify big linguistic models and AI agent really suggest.”
Trial development
What does this mean in practice? Paskalev says Logicstar performs an analysis of each application in which its technology is developed – using “classic IT methods” – to build a “knowledge basis”. This gives the AI agent a comprehensive map of the inputs and expense of the software. How the variables are linked to the functions. and any other connections and dependencies, etc.
Then, for each error presented, the AI agent is able to determine which parts of the application are affected – allowing Logicstar to limit the functions that must be simulated in order to test possible ratings of possible corrections.
Per Paskalev, this “minimized execution environment” allows AI agent to perform “thousands” tests aimed at reproduction .
Confirms that the actual bug fixes come from LLMS. But because the Logicstar platform allows for this “very fast executive environment”, AI agents can operate on a scale to separate the wheat from the chaff, as it was, and serve its users with a shortcut at the best of which LLMS can offer.
‘What we see is [LLMs are] Excellent for original, testing things, etc. but not absolutely great for [code] Production, commercial applications. I think we are far from there, and this is what our platform delivers, “he argued.” In order to be able to export these models of the models today, we can really export the commercial value and save time for the developers to really focus on the important things ”.
Businesses are going to be Logicstar’s original goal. His “silicon agents” are meant to work with the DEV corporate groups, although in a fraction of the salary required to hire a human developer, handling a series of applications maintenance tasks and release of engineering talents for more creative and/or provocative work. (Or, at least, until llms and ai agents become much more capable.)
While the starting tar shows a “fully autonomous” application maintenance capacity, Paskalev confirms that the platform will allow people to review (and oversee) the corrections of AI agents. Thus, trust can be – and must be – earned first.
“The accuracy delivered by a human developer ranges between 80 to 90%. Our target [for our AI agents] It must be right there, ”he adds.
It is still the first days for Logicstar: an alpha version of its technology is in tests with a number of non -announced companies that Paskalev is referred to as “design partners”. Currently, technology only supports Python – but extensions to Typescript, Javascript and Java are charged as “Coming Soon”.
‘The main objective [with the pre-seed funding] It is to show that technology works with our design partners – focusing on Python, “adds Paskalev.” We’ve already spent a year on it, and we have many opportunities to really expand. And that’s why we try to focus it first, to show the value in one case. ”
The increase in launch pre-section was driven by the European company VC Northzone, with angels investors from Deepmind, Fleet, Sequoia Scouts, Snyk and Spotify also participating in the round.
In a statement, Michiel Kotting, a Northzone associate, said: “The generation of code led by AI is still in its early stages, but the productivity profits we already see is revolutionary. The ability of this technology to rationalize the processes Development, to reduce costs and accelerate innovation is huge. software. “
Logicstar works a waiting list For potential customers wishing to express interest in accessing early. He told us that a Beta version is designed for later this year.
