- Klarent has attracted $8 million in seed financing, with Mosaic acting as the lead and Moonfire also participating to address outsourced quality assurance.
- Instead of using model calls while the software is running, Klarent has AI agents that write and reuse test code.
- A large portion of the funds will be used to hire sales staff in the United States and the United Kingdom. The company currently has 25 employees.
Asheem Panakkat, originally from India, was a software engineer and engineering manager at Google, where he worked on products such as Maps, Search, Shopping, and Lens. In 2023, he and his colleague Momchil Ivanov left Google to found Klarent.
At first, they developed evaluation tools for large language models, but later concluded that software testing for AI-generated code presented a greater opportunity. This shift enabled Klarent, based in Zurich, to raise $8 million in seed funding, with Mosaic Ventures leading the investment and Moonfire Ventures also participating.
The list of angel investors includes Harris Barton, AngelInvest Ventures, and Jürgen Galler, referred to in the announcement as a former Google director, along with others.
Speaking to Tech Funding News, Panakkat said, “Quality assurance is a $35 billion business, give or take, and most of it is carried out manually. Klarent aims to reduce labour costs.”
“Our first step was to develop evaluations for LLMs, but we soon realised that the greater opportunity lay in QA. Since AI is going to write more of the world’s code, testing will cease to be a cost centre and will instead become the factor that decides whether any of that code can be trusted,” adds Panakkat.
The tests are written by AI, but it does not run them
According to Klarent, AI should write tests rather than run them. Panakkat explains, “Many of the startups in this area are taking an approach which I would refer to as probabilistic, in which agents make model calls as the software runs. It’s very impressive on a small scale, but it doesn’t actually work when it comes to enterprise scale.”
Klarent reduces costs by reusing code the models develop through caching, as Panakkat stated. One customer carries out approximately 3,500 tests each day.
The company calls this “deterministic autonomy” and, as the announcement states, a human-in-the-loop service checks each result. The writing and repair processes still involve a model.
On competitors, Panakkat said Octomind, which raised $4.8 million in seed funding, could not operate this at scale, and that Tricentis is working to add an agentic feature to its existing product.
QA Wolf and Momentic have secured larger funding rounds in the United States, and TFN has also mentioned LambdaTest‘s $45 million financing and Sweden’s Endform.
Challenging outsourcing, with pricing still to be clarified
Chandar Lal, partner at Mosaic, identified traditional outsourcing as its main competitor: “In the old days you depended on people, usually those who were outsourced to a low-cost country, to produce a large number of test scripts and then click a large number of buttons. It was a brute-force approach, and there is a huge service market in this area bringing in hundreds of billions of revenue.”
He adds, “A great deal of the white-collar professional services market will be completely wiped out, and software testing through outsourcing is a perfect example of this, even though no one has worked out how to deal with it.”
As Panakkat points out, at JD Sports the agents carry out processes which move “from placing the order on the website through to the order management system and then on to the delivery systems.” He adds that in-house software development engineers doing test work could not achieve that level of automation.
Why Mosaic invested and how the funds will be used
Panakkat believes a large services industry was waiting to be disrupted and that this represented a significant opportunity, adding that Chandar and his team spoke their language.
“At Google, both Asheem and Momo noticed this phenomenon before the rest of the industry did. What convinced us was their firm belief in determinism, using AI to write tests rather than run them, which is why businesses are now placing their trust in Klarent for use in production,” notes Lal.
The funds will be used to expand in the United States and for engineering. Panakkat said most of the money will go to hiring in the US, expanding the sales team in the US and the UK, and, to a lesser degree, engineering. The company currently has 25 employees.
Looking ahead, Panakkat expects “a network of agents that are making software and a network of agents that are reviewing or quality-assuring that software.”
