Summary

• Mosaic Ventures led Klarent’s €7.13 million seed round.
• The Zurich startup was previously called fore ai and was founded by Asheem Panakkat and Momchil Ivanov.
• Its stated design uses AI to write tests, with deterministic execution and a human review layer.

Zurich-based Klarent has raised €7.13 million in seed financing for a testing platform that uses artificial intelligence to write software tests while keeping their execution repeatable. Mosaic Ventures led the round, with Moonfire Ventures, AngelInvest Ventures and angel investors taking part. The financing will support product development and plans to expand into the United States.

The company was previously known as fore ai and was founded in 2023 by former Google engineers Asheem Panakkat and Momchil Ivanov. It describes a division between the less predictable work of generating tests with a language model and the controlled execution needed to decide whether an application behaves as expected. Its investor announcement and contemporary funding reports describe human review as part of that process.

That separation addresses a practical consequence of AI-assisted programming. Developers can produce code, interfaces and integrations more quickly, but an increased volume of changes does not establish that software is safe to release. New features can affect authentication, databases or established customer journeys in ways that a demonstration of the new functionality does not reveal.

Tests must be able to detect failures and give engineers evidence they can reproduce. When a test produces different outcomes without a corresponding change in the application, teams must spend time establishing whether the defect lies in the software, the test or the environment. Klarent’s proposition is to use language models where they assist test authoring, then avoid relying on model-generated judgements every time those tests run.

Deterministic execution is a meaningful engineering property, but it is not sufficient proof of test quality. A repeatable test can consistently check the wrong requirement. If test data do not resemble production conditions, or if the test overlooks an important integration, a passing result can create false confidence. The architecture still needs appropriate coverage and careful selection of the behaviours being verified.

Large enterprise applications complicate that task because software rarely operates alone. A business workflow may move through a user interface, identity provider, several internal services and an external payment or data system. Tests that only inspect individual components can miss defects in the interactions between them. Conversely, broad end-to-end tests can be slow and fragile when environments or dependent services change.

Adoption therefore depends on how Klarent works with existing repositories, continuous integration pipelines and release approval processes. Engineering teams need to understand what the AI has generated, how tests are reviewed, when they execute and what evidence is retained after a failure. The company also describes an outcome-based commercial element, but reported pricing language is not a public, independently audited measure of results.

Automated testing can itself become a maintenance burden. As software changes, teams must update tests, remove obsolete scenarios and investigate false alarms. Generating more scripts is not necessarily useful if those scripts duplicate coverage or produce alerts that developers cease to trust. The relevant productivity measure is the effort required to detect and resolve genuine defects, balanced against the cost of keeping tests current.

Investors see an opportunity in the gap between faster code production and slower assurance, yet that gap also attracts established software vendors and competing AI testing startups. Klarent will need to demonstrate how its particular method performs across customer environments rather than rely on architectural differentiation alone. Contemporary reports describe enterprise interest, but do not provide a controlled benchmark demonstrating lower defect rates.

Klarent’s approach also requires teams to decide who owns the test specification. Product managers may define the expected business outcome, engineers may implement automated checks, and quality specialists may review coverage. AI-generated tests can accelerate the intermediate work without settling disagreements about what the software should do. A repeatable execution result should remain traceable to a requirement and to the version of the application examined. Otherwise, automation could increase the volume of passing checks without improving release decisions.

The €7.13 million round gives the business resources to pursue engineering work and US expansion. Evidence of commercial progress will come from durable customer deployments, reproducible testing results and measurable changes in release quality or maintenance effort. For now, deterministic execution is a verifiable description of the proposed method, while better outcomes remain a proposition to be tested.