Swiss startup start Logic He is determined to join the game of artificial intelligence agent. The startup obtained the establishment of the summer of 2024 $ 3 million as a financing before a seed to bring tools to the developer market, which can carry out independent maintenance of software applications, instead of using the most typical artificial intelligence agent for joint development.
Logicstar CEO and co-founder Boris Paskalev suggests that AI agents may end up in partnership with code development agents-for example, for example, CONTION LABS-in winning work.
Code Fidelity is a problem in creating and publishing AI agents for employees, as is the case for human developers, Logicstar wants to do a few to distort the development wheel by picking up and fixing errors automatically wherever they are in a symbol published.
As is the case, Paskalev suggested that “even the best models and agents” there are unable to solve the majority of errors that are presented with it – hence the team was spying on the opportunity to start the start of the designated Amnesty International to improve these difficulties and provide them with a lessen application of the application.
To this end, they build large linguistic models (LLMS)-such as the GPT from Openai or even Deepseek in China-follow a result of the model for their platform. This allows Logicstar to decrease in different LLMS and increase the benefit of AI’s agents, based on the basic model that works better to solve a specific symbol problem.
Paskalev claims that the founding team has technical knowledge and knowledge of the field to build a platform that can solve programming problems that can challenge or work to work alone. They also have a former success in organizing project to refer to: The former Code Review, Deepcode, has sold to SNYK in SNYK in September 2020.
“Initially we were thinking about building a large language language model,” he told Techcrunch: “Then we realized that this will quickly become a commodity … Now we are building on the assumption that all large language models are present. Assuming that there are already some decent agents (AI) for the symbol, how can we extract the maximum value of work from them?”
He said that the idea based on the team’s understanding of how to analyze software applications. “Gather this with large language models – then focus on grounding and checking what is proposed by large language models and artificial intelligence agents already.”
The development of the test
What does this mean in practice? Paskalev says that Logicstar analyzes for every application whose technology is published – using “classic computer science methods” – in order to build a “knowledge base”. This gives artificial intelligence agent a comprehensive map of the program’s inputs and outputs; How variables are related to jobs; Any bonds, other consequences, etc.
After that, for every mistake it is presented with it, the artificial intelligence factor is able to identify the parts of the affected application – allowing Logicstar to narrow the jobs that need to simulate in order to test dozens of possible repairs.
For each PASKALEV, the “Implementation Environment” provides the artificial intelligence factor to operate “thousands” of tests aimed at reproducing errors to determine a “failed test”, and through the “test-based development” approach-ultimately to fix that stick.
It confirms that actual errors are obtained from LLMS. But since the Logicstar platform enables this “very fast executive environment” AI agents can work widely to separate wheat from the peel, as it was, and its users in a nutshell serve the best LLMS can offer.
“What we see is (LLMS) is great for initial models, testing things, etc., but it is not absolutely great for production (code), commercial applications. I think we are far from there, and this is what our platform delivers.” “In order to be able to extract these capabilities from the models today, we can actually extract the commercial value safely and save time in reality for developers to really focus on important things.”
Institutions have been appointed to be the initial goal of logic. Silicon agents aim to work alongside the DeV teams for companies, although it is on a small part of the salary required to employ a human developer, and to deal with a set of application maintenance tasks and free engineering talents for more creativity and/or challenge work. (Or, at least, at least until LLMS and artificial intelligence agents get more capable.)
Although the startup stadium describes the “completely independent” application capacity, Paskalev confirms that the statute will allow human developers to review (and involve them) in repairs. So confidence can be – and it must be – earning first.
“The accuracy offered by a human developer ranges from 80 to 90 %. Our goal (to our artificial intelligence agents) is to be exactly there.
There are still early days for Logicstar: Alfa version of its technology in the test with a number of unannounced companies that Paskalev refers to as “Design Partners”. The technology currently supports Python only – but expansions are described on Typescript, Javascript and Java as “coming soon”.
“The main goal (with financing before the seed) is to show technological works with our design partners-with focus on Bethon,” he adds. “We have already spent a year, and we have a lot of opportunities to actually expand. For this reason we try to focus first, to show the value in one case.”
European Northzone has led to the buyer of the startup, as it has joined investors from DeepMind, Fleet, Sequoia Scouts, SNYK and Spotify also to the tour.
In a statement, Michelle Cotting, a partner in North Basun, said: “The generation of the code that is moved by artificial intelligence in its early stages, but the productivity gains that we already see are still revolutionary. The possibility of this technology to simplify development processes, reduce costs, and accelerate innovation is enormous. The extensive technical expertise of the team and its registration places them in providing real and effective results.
Logicstar works waiting list For potential customers who want to express their interest in early access. He told us that a demo version was planned later this year.
