Klarent raises $8M: AI writes the tests, the tests run without AI
Klarent's $8M seed backs AI software testing where LLMs write test code but never run at test time. That design is the cheap, repeatable QA small teams need.
Zurich-based Klarent raised an $8 million seed round for AI software testing with one rule we like: the model writes the test, then gets out of the way. Its agents author and repair test code with an LLM, but the tests themselves run with zero model calls. Same input, same result, every run.
What actually happened
Per Tech Funding News, Mosaic Ventures led the round, with Moonfire Ventures and angels including Harris Barton, AngelInvest Ventures and former Google director Jürgen Galler. Founders Asheem Panakkat and Momchil Ivanov left Google in 2023; Panakkat worked on Maps, Search, Shopping and Lens. The company has 25 employees and started out building LLM evaluation tools before pivoting to QA.
Klarent calls the approach "deterministic autonomy." Agents generate and fix test code, the generated code is cached and reused to keep compute costs down, and a human-in-the-loop service checks each result. Panakkat contrasts this with rivals whose agents call a model while the tests run, which he says doesn't scale. One customer runs about 3,500 tests a day; JD Sports uses it to test flows from web order through order management to delivery. Most of the money goes to US hiring and US/UK sales.
Why it matters for your business
We have run the other version of this. An agent that "clicks through checkout" with a model on every step is slow, costs tokens on every run, and fails differently on Tuesday than on Monday. A flaky test is worse than no test, because the team learns to ignore red.
Use AI at authoring time, not at runtime. That is the pattern worth stealing, whether or not you buy Klarent. Have Claude or GPT write Playwright tests for your checkout, account and refund flows. Review them. Commit them. Run them in CI for free, forever.
Spend model calls only on change. When the UI shifts and a selector breaks, let the agent propose a fix as a pull request. A human approves. The test suite stays code you own and can read.
Count the bill per run. If your testing vendor charges per execution because a model runs every step, your QA cost scales with your deploy frequency. That is backwards.
Key takeaways
- Klarent raised an $8M seed led by Mosaic Ventures for AI-driven software testing
- LLMs write and repair test code; tests execute with no model calls, and code is cached for reuse
- A human-in-the-loop service reviews results; one customer runs about 3,500 tests a day
- The pattern for small teams: AI authors Playwright tests, humans review, CI runs them deterministically
- Avoid QA tools that bill model calls on every test run
Checkout breaks and you find out from a customer? We write end-to-end test suites for your money paths, AI-drafted, human-reviewed, and running in your CI at no per-run cost. See what we build, or price what a broken checkout costs you.
Sources: Tech Funding News, EU.VC.
- #ai-software-testing
- #klarent
- #qa-automation
- #deterministic-tests
- #funding
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Sakana AI review agent: 70% of planted errors, 16% of real ones
Sakana AI's Multi-Layered Review catches over 70% of planted errors but 16% of real ones in withdrawn papers. Test AI review tools on your own mistakes.
Read itBitwarden app store builds go commercial: test your export
Bitwarden's app store and download builds switch to a commercial license; GPLv3 code stays on GitHub. What changes for your team and why to test your export.
Read it