In software, the most expensive line of code is the one you have to write twice. The second most expensive is the one you never wrote at all, because a bug slipped through. For Asheem Panakkat and Momchil Ivanov, two engineers who spent over a decade at Google, the solution to this old problem looks less like a new programming language and more like a traffic controller for large language models.
Their Zurich-based startup, Klarent, sells what it calls an autonomous QA platform. The pitch is straightforward: describe what a feature should do in plain English, and an AI agent will write the test code, execute it deterministically, and maintain it. The clever bit, and the one that secured an $8 million seed round last October, is what it doesn’t promise. Klarent is not selling a black-box AI that whimsically decides if your app works. It is selling a system where AI writes the tests, but a human engineer remains the final authority, verifying outcomes and signing off on releases [Klarent product description]. It’s a bet on augmentation, not replacement.
The Wedge: Determinism Over Magic
The classic trap for any AI testing tool is flakiness. An LLM might write a brilliant test one day and a nonsensical one the next, or it might interpret the same requirement differently on each run. For an enterprise like JD Sports, NZZ, or Sixt,three of Klarent’s named customers,that kind of unpredictability is a non-starter [The SaaS News, October 2026].
Klarent’s technical wedge is to split the workflow. First, an LLM acts as a code author, translating natural language into executable test scripts. Then, the platform takes over, running those scripts in a repeatable, deterministic environment. The final, crucial layer is a human-in-the-loop (HITL) gate where an engineer reviews the results. This structure aims to capture the speed of AI-assisted creation while preserving the reliability and accountability that enterprise development teams require [Klarent product description]. It’s a recognition that while AI can generate enormous volumes of test code, someone still needs to own the risk.
The Team and the Check
The founders’ Google pedigree is the cornerstone of Klarent’s story. Panakkat, the CEO, led teams on products like Google Shopping and Lens. Ivanov, the CTO, was a staff software engineer focused on information retrieval [Klarent about page]. They are not career QA specialists, which may be the point. They are product builders who understand the scale and complexity of modern software release cycles, and the bottlenecks that form around manual testing.
That background helped them assemble a seed round that suggests investors see a path beyond a niche tool. The $8 million (also reported as €7.13 million) round was led by London-based Mosaic Ventures, with participation from Moonfire Ventures, AngelInvest Ventures, and angels including former Google director Jürgen Galler [Klarent newsroom][The SaaS News, October 2026]. The stated use of funds is expansion into the U.S. market, engineering hires, and extending the platform’s reach. Klarent started with web testing but has since added support for native Android and iOS applications, a necessary expansion for its target enterprise clients [Klarent newsroom].
| Founder | Role | Key Background |
|---|---|---|
| Asheem Panakkat | Co-Founder & CEO | 12 years at Google (Shopping, Lens); MBA, Indian School of Business [Klarent about page][RocketReach]. |
| Momchil Ivanov | Co-Founder & CTO | Staff Software Engineer at Google, information retrieval [Klarent about page]. |
Traction and the Microsoft Play
Public traction metrics are scarce, but the company has signaled early enterprise adoption and a savvy distribution move. Beyond the customer names, Klarent has made its platform available through the Microsoft Marketplace, a channel that can provide credibility and streamlined procurement for other Microsoft-centric enterprises [Klarent newsroom]. A former business development employee also noted developing a sales pipeline worth over €300,000 in early 2026 [Simon Hecker - fore ai | LinkedIn, retrieved 2026].
The company, which rebranded from ‘fore ai’ to Klarent, appears to be positioning itself as a platform rather than a point solution. The agentic testing approach it promotes,where AI agents handle the creation and maintenance of tests,is a step beyond record-and-playback tools, aiming to integrate into the entire software development lifecycle [MoTaverse, September 2026].
The Human Bottleneck
For all its promise, Klarent’s chosen path introduces its own set of constraints. The human-in-the-loop model is its key differentiator, but it is also a potential ceiling on scalability and speed. Academic and industry notes on HITL systems highlight that they require continuous monitoring, clear oversight principles, and metrics to assess the AI’s output, all of which demand time and attention from already-busy engineers [Devoteam, Unknown][IAPP, Unknown]. The very gate that ensures quality could become a bottleneck if not managed exquisitely well.
The competitive landscape is another open question. While no direct competitors are named in the available sources, the market for test automation is crowded, ranging from legacy suites to a new wave of AI-native tools. Klarent’s bet is that its combination of deterministic execution, cross-platform support (web, iOS, Android), and the HITL safety rail will carve out a defensible position with risk-averse, quality-conscious enterprises.
The Unit Economics of Trust
The real test for Klarent won't be whether its AI can write a test, but whether its system can save more engineering hours than it consumes. Imagine a mid-sized engineering team that spends 20 hours a week on manual test creation and maintenance. If Klarent can cut that to 5 hours of review and oversight, that’s a 15-hour net saving per week. At a blended engineer cost of, say, $100 per hour, that’s $1,500 weekly, or about $78,000 annually, in recovered capacity for a single team. The platform’s price point needs to sit comfortably under that recovered value, across dozens of teams, to make the ROI unambiguous.
Klarent’s ultimate incumbent isn’t another AI startup. It’s the entrenched habit of manual scripting and the cautious inertia of engineering leaders who would rather trust a known, slow process than a fast, unfamiliar one. To win, Klarent must prove its agents are not just clever, but trustworthy,and that its human-in-the-loop is a lever for scale, not a brake.
Sources
- [Klarent product description] About Klarent | https://klarent.ai/about/
- [The SaaS News, October 2026] Klarent Raises €7.13M Seed | https://www.thesaasnews.com/news/klarent-raises-7-13m-seed/
- [Klarent newsroom] Klarent News | https://klarent.ai/news/
- [Klarent about page] Team | https://klarent.ai/about/
- [MoTaverse, September 2026] Klarent joins us at MoTaCon 2026 | https://www.ministryoftesting.com/moments/klarent-joins-us-at-motacon-2026
- [Simon Hecker - fore ai | LinkedIn, retrieved 2026] Simon Hecker LinkedIn Profile | https://www.linkedin.com/in/simon-hecker-35442a43/
- [RocketReach] Asheem Panakkat Profile | https://rocketreach.co/
- [Devoteam] Human-in-the-loop systems | https://www.devoteam.com/expertise/artificial-intelligence/human-in-the-loop/
- [IAPP] Solving AI risks with HITL | https://iapp.org/news/a/solving-ai-risks-with-human-in-the-loop-systems/