An engineering team in San Francisco pushes code at 4:47 p.m. on a Friday. Nobody flinches. Nobody pages the on-call. The regression suite, written in plain English by a product manager, fans out across hundreds of parallel browsers. Twelve minutes later, the green light. Welcome to a small, weird corner of the software industry that Rainforest QA quietly rebuilt.
A testing company that finally stopped sounding like one
Rainforest QA sells an AI-powered, no-code platform for end-to-end testing. That is the boring sentence. The interesting sentence is that they let you write a regression test the way you would describe it to a new hire - in English - and then run it on thousands of browsers at once. The result looks less like QA and more like a utility. Press play. Watch numbers go up.
The company today employs roughly 120 people, mostly distributed, mostly engineers, and serves a customer list dotted with the kind of mid-market SaaS companies that have outgrown manual testing but cannot stomach the cost of a Selenium maintenance team. Mattermost. Karbon. Domino. MindBodyGreen. The names suggest the pattern: software companies who would rather spend cycles on product than on test scripts that broke when somebody renamed a button.
QA should help teams move faster - not slow them down.
Above: A sentence that nobody at any QA vendor would have written in 2009.
The QA tax nobody wanted to pay
For most of the 2010s, shipping web software meant accepting one of two bad deals. You could hire a QA team, watch it grow at the same rate as your codebase, and discover that humans clicking through checkout flows do not scale. Or you could buy into the religion of test automation, write Selenium scripts in JavaScript, and learn that automated tests rot at roughly the same rate you write them. Either way you paid the QA tax.
Rainforest's founders, Fred Stevens-Smith and Russell Smith - same surname, not related, a confusion they have learned to live with - noticed that the tax was structural. Tests were brittle because they were code, and code only stays alive when somebody pays to maintain it. Testing teams were expensive because they were teams. The whole apparatus existed to slow you down at the worst possible moment: right before you shipped.
We are building the AWS for QA, so that software companies don't have to build a QA team.
From a crowd of humans to a swarm of AI
Rainforest's first incarnation was, frankly, the strangest pitch in the YC Winter 2012 batch: an on-demand crowd of human testers who would click through your app on real machines for a few dollars a head. It worked, in a brutalist sort of way. Customers loved the consistency. The crowd loved the work. The unit economics, however, were exactly what you'd expect when your product had a labor input.
So the founders did the thing that founders only do when they have run out of other ideas: they re-pointed the company. The crowd became training data. Years of humans clicking through the same checkout flows taught Rainforest's models how to drive a browser themselves. By the late 2010s the platform had shifted from people-powered to AI-powered, and the company began describing itself, somewhat shyly, as the AWS for QA.
The pivot is the sort of thing that looks inevitable in hindsight and was almost certainly not. Pivots rarely are.
A quarter of all testing was successfully automated by the platform after the model began driving the browser.
Translation: a meaningful fraction of regression tests stopped requiring a human or a script. They just ran.
What you get when you press play
The Rainforest product, stripped to its essentials, does three things. It lets you author tests in natural language, with no SDK and no code. It runs them in parallel, on a fleet of real browsers, in a few minutes. And it heals them, more or less automatically, when your engineers rename a div or move a button two pixels to the left.
There is also a visual layer test, which catches the kind of CSS regression that humans tend to miss until a customer screenshots it on Twitter. There is a CLI and an API, which means the whole thing slots into a CI pipeline without ceremony. And there is, more recently, a self-healing layer that uses AI to figure out what you meant when the page you wrote a test against no longer exists in quite the same shape.
A short history of Rainforest
Twelve years, two business models, one stubborn idea.
The case for skepticism, answered with numbers
Test automation is a sector littered with the corpses of products that promised faster shipping and delivered slower flakes. So the right response to any Rainforest claim is, reasonably, to ask for receipts. Here are the receipts: over 42 million tests run in production. Roughly 120 employees on the payroll. A customer base of more than 10,000 startups and product teams. Hundreds of thousands of bugs caught before they ever made it to a customer's browser.
Where Rainforest moves the needle
Source: directional figures synthesized from publicly reported customer outcomes. Your mileage, as ever, will vary.
The customer logos do most of the rhetorical heavy lifting. Mattermost runs Rainforest. So does Karbon, the accounting workflow tool, and Domino, the data science platform. None of them are companies that would tolerate a flaky test rig for very long. The platform's continued use by teams who could plausibly build their own automation says something a brochure cannot.
Hundreds of thousands of bugs caught before they reached a customer.
QA, but as utility
The Rainforest mission, properly understood, is not about testing at all. It is about taking a function that every software company is forced to perform and turning it into a thing you pay for by the minute. Electricity used to be on-prem. So did compute. So did, briefly and disastrously, email. Rainforest's bet is that QA belongs on the same list.
If that bet pays off, the cultural shift is the interesting bit. Engineering teams stop arguing about who owns the test suite. Product managers learn to write their own checks. Releases stop carrying the implicit risk premium that any sufficiently large QA backlog generates. The company has been making this argument for over a decade, and the world has slowly come around to it.
An AI-shaped problem with an AI-shaped answer
The next phase, the company is happy to admit, looks unlike the last. AI-generated code is being written faster than humans can review it, which means the QA bottleneck just moved. The volume of code that needs testing is going up. The number of humans available to test it is not. Whatever you think of generative AI, the second-order effect on QA is uncontroversial: somebody, or something, needs to run more tests, faster, with less maintenance.
Rainforest, having pivoted once from crowd to automation, is rather well-positioned for this. The platform already speaks plain English, already runs in parallel, already self-heals. The frontier is what happens when AI-generated tests meet AI-driven execution meet AI-written code. The answer to that question is being prototyped in Rainforest's San Francisco office, and is the reason a Series B from 2018 has not yet been outgrown.
It is also the reason that, at 4:47 p.m. on a Friday, in a startup somewhere, an engineer is shipping code, the regression suite is running itself, and nobody is paging the on-call. The test rig was once the loudest part of the room. Now it is the quietest. That is the change.