Loopwright announced a $54 million Series A on Sunday led by Accel with participation from Elevation Capital and a group of angels including several engineering leaders from Atlassian and Freshworks. The Bengaluru company builds test generation software that works from production session recordings: it observes what real users actually do, clusters those behaviors into representative flows, and generates end-to-end tests covering the paths that matter by usage rather than by whatever the team remembered to write. Roughly 90 customers are paying, including two publicly traded software companies, and annual recurring revenue crossed $8 million in June.
The problem with how tests get written
End-to-end test suites are written from specifications and from bugs that already happened. They cover the flows an engineer imagined and the ones that broke. What they systematically miss is the way real users navigate, which involves backing up mid-flow, opening things in new tabs, leaving a form for twenty minutes and returning, and combinations of state nobody designed for. Those paths produce a meaningful share of production bugs and almost none of the test coverage.
The other problem is maintenance. A large end-to-end suite is expensive to keep running, breaks constantly on interface changes, and accumulates flaky tests that teams eventually learn to ignore. Surveys consistently find engineers distrust their own end-to-end results, and a suite nobody trusts is worse than no suite because it consumes time while providing false assurance.
What Loopwright does
A lightweight instrumentation library records anonymized interaction sequences in production: which elements were interacted with, in what order, with what timing, and what the resulting application state was. Personally identifiable content is redacted at the client before transmission, which the company documents in detail because it is the first question every prospect asks. The service clusters sequences into flows, weights them by frequency and by business value signals the customer configures, and generates Playwright tests.
The generated tests use semantic selectors derived from accessibility attributes and text content rather than CSS paths, which is what makes them survive interface changes. When a test does break, the system correlates the failure against the production recordings that produced it and proposes a repair, which a human approves. Chief executive Rohan Iyer said the acceptance rate on proposed repairs runs around 70 percent, and that getting it above that is where most of the engineering effort goes.
The story is rarely the launch. It is what breaks, what ships, and who owns the mess at 2 a.m.
The evidence it works
A publicly traded customer, which Loopwright would name only as a North American human resources software company, shared aggregate results. After six months, its generated suite covered 340 flows against 190 hand-written tests, and in that period caught 23 regressions that the existing suite missed. Fourteen of those involved state combinations the team described as things nobody would have thought to test. Flake rate on generated tests ran at 2.1 percent against 9 percent on the hand-written suite.
The counter-evidence is that generated tests encode current behavior rather than intended behavior, so a bug that has existed long enough to appear in production recordings becomes a test asserting the bug. Iyer acknowledged this directly and said the product flags flows with anomalous outcomes for human review rather than pretending to solve it. Several customers described that as the main ongoing operational cost.
Competition and category
Mabl, Testim, and Functionize all sell AI-assisted test authoring and maintenance, and all approach it from the authoring side: describe a test in natural language, get a script. That is a different product solving a different problem, and it does not address coverage of unimagined paths. Playwright itself has a recorder, which produces a test from one manual session and has no clustering or maintenance.
The closest competitor conceptually is a feature inside several session replay products, notably FullStory and LogRocket, which have both discussed generating tests from recordings. Neither has shipped it as a serious product. Accel partner Prayank Swaroop, who joins the board, said the thesis was that this is a full product rather than a feature, because the hard part is the test maintenance loop rather than the initial generation.
The risks
Instrumenting production to record user behavior is a privacy exposure and a sales obstacle, particularly in Europe and particularly for anyone in healthcare or finance. Loopwright's redaction happens client-side and its architecture keeps raw recordings in the customer's own storage, with only derived flow structures reaching Loopwright's systems. That design was clearly built for the objection and it will still lose deals at organizations that will not instrument production at all.
The other risk is that the major testing frameworks absorb the idea. Microsoft maintains Playwright and has both the data access through Azure and the incentive. Iyer's answer is that the value is in the maintenance loop and the accumulated repair models, which improve with usage across customers, and that a framework vendor building it would be starting from zero. That is the standard answer and it is more convincing here than in most cases, because the loop genuinely does compound. The counterweight is that Microsoft does not need to match the loop's quality to win, only to be adequate and free, which is the argument that has ended a long list of otherwise excellent developer tool companies. Iyer's plan is to be far enough ahead on the maintenance experience that adequate is visibly not good enough.
Skarvonix will keep following this beat with reporting grounded in how systems behave outside the launch keynote.
- Funding
