One tweet, a two-letter package name, a YC company: e2e hit 8,500 stars in ten days, but it's not a path an individual developer can copy

Two young Polish developers with React Native backgrounds, and Oct 1 posted a tweet turning their company's paid testing platform into an open-source framework. A month later the repo had 13k stars, npm was getting about 100k downloads a day, and the company had raised a 8M-dollar seed round. The product idea is worth learning from, and so is the launch tactic. But if you want to build a similar project to chase stars, first look at what they had in hand when they started.

Oct 1 tweet at 3 p.m.

2026 Oct 1 15:04 (UTC), Oskar Kwaśniewski (@o_kwasniewski on X) posted a tweet:

Introducing e2e — The agentic testing framework for any app ~/ npx e2e init → Fully Open Source → Mix deterministic and agentic APIs → Supports web, mobile, and more → Bring your own agent and infrastructure → Run locally or in CI

It came with an image. As of 11-06, the tweet has 1.2M views, 4.9K likes, 629 reposts, and 8.5K bookmarks, more than the likes. Lots of bookmarks mean people saved it as a tool to use later, not just tapped a quick like.

The npm download curve lines up with this tweet: the package e2e was getting about 800 downloads a day in late September, 10-03 reached 19,636,10-12 by 67,327,11-03 and 118,763 (npm download stats). The GitHub repo was created in Jul 22, but made public in Oct 1 (likely private before that); by 10-10 it had 8,562 stars, meaning about 8,500 within less than ten days of going public.

Another comparison shows the point: 10-12, someone posted a Show HN, and it only got 9 points. The product alone spreading within developer communities doesn't reach that scale.

What it is

A TypeScript framework for end-to-end testing of web, iOS and Android apps. What's distinctive is that you can write both kinds of steps in the same test. The first code block in the README looks like this:

``ts test('a member upgrades to Pro', async ({ app, agent, screen }) => { await app.open('/settings/billing'); await agent.act('upgrade the workspace to the Pro plan'); await agent.assert('the invoice preview shows a prorated amount'); await expect(screen.getByRole('status')).toContainText('Pro'); }); ``

The two middle lines are natural language, which the AI clicks through and judges; the last line is traditional code, using locators and assertions for precise checks. One key design point: steps the AI took are recorded once a later assertion verifies them, and replayed next time without calling the model, and the AI only walks through them again when the app's UI changes. Tests with no AI steps don't need a model at all.

Developers' concerns about "AI-written tests" mainly come down to two things: they're flaky when run, and calling the model every time is expensive. The mixed style plus cached replay addresses exactly those two.

Several other details are aimed at "letting coding agents write and run tests themselves": you can use your Claude Code or Codex subscription directly to run the AI steps; all the docs are bundled in the npm package (node_modules/e2e/docs) so agents can read them offline; and npx e2e init spins up a project in one command.

How it got 8,500 stars in ten days

The founders were already known. Both founders come from Callstack, a well-known React Native consultancy in Poland. Szymon Rybczak maintained the React Native Community CLI; Oskar built react-native-visionos, the React Native port for Apple Vision Pro, and often speaks at RN conferences. So when the tweet went out, the first readers were RN and Expo developers. And e2e supports mobile, which hit exactly that crowd: mobile testing options had mostly been older tools like Detox and Maestro.

YC and investors backed it. They were in YC P26, and their June Launch HN got 132 points; in Sep 15 they announced a 1.2M-dollar pre-seed; in Sep 18 YC partner Pete Koomen tweeted in support, saying they had built the mobile app for Vercel v0.

Partner vendors reposted it. The README lists a row of official integration packages: TestMu AI (formerly LambdaTest), BrowserStack, Kernel, Lightpanda, smol machines, Limrun, Expo EAS. Each added integration means another vendor who writes their own blog and reposts it. TestMu's official blog even ran a post titled "e2e by TesterArmy".

They got a great name. On npm, e2e was an old package name that someone had registered 2014 years earlier; they got it, and also got e2e.com. "e2e" is a word developers say every day, easy to remember, and easy to find when searching "e2e testing".

The company bet all its resources on this one repo. The original closed-source TesterArmy platform stopped accepting new signups, shut down in 2027 Jan 15, and sent all new users to the open-source e2e. In Nov 3 the company simply renamed itself e2e and announced a 8M-dollar seed round. The star curve shows this jump: from 11-03 to 11-05 new stars a day were 149, 230, 310, well above the 50 to 120 per day the week before (GitHub daily star counts).

One more move is worth noting: in late October they opened a new scenario example repo on GitHub almost every day: e2e-stripe, e2e-paypal, e2e-revenuecat, e2e-clerk, e2e-shopify, e2e-biometrics, e2e-deep-links and more. This is capturing long-tail searches like "how to test Stripe payments" and "how to test Face ID."

Why it keeps people

The tweet explains the first ten days of the spike, but not what came after: npm daily downloads climbed to roughly 100k a day, and the company itself says 1.5M downloads in the first month and 60 external contributors. Last week @e2e-dev/web got 627k downloads, Midscene (@midscene/web), the best-known open-source project in the same space, got 73k, and Shortest got 11k (npm, last-week downloads). People are clearly using it more than its peers. How many of the downloads on the main package e2e come from CI reinstalling it over and over, and how many from people chasing the old package name, you can't tell from the outside, so discount that number.

What keeps people around comes down to the product differences above: deterministic and AI steps mixed in one test, cached replay after verification, one API for web and mobile, and the model is your choice.

Who it's competing with

(GitHub stars, early November)

Copycats have already appeared: Oct 21 built Dánwò (carbonmeai/danwo), whose description calls itself "the most complete open-source e2e testing framework for the agentic era". 10-23 also posted a Show HN, and at the time of research it had only 18 stars. Copying the pitch won't get you their stars.

If it were you

Bottom line: Going head-on with an "AI testing framework" to fight it for stars isn't worth it. It took off because of its founders' standing in the community, YC, two funding rounds, a lineup of partner vendors, and the full resources of a commercial platform it shut down. Individual developers don't get any of that. It's also already been widely covered, and the space has Microsoft, ByteDance, and a pile of YC companies in it. But several of its design and launch tactics are very useful for other projects.

Who it suits: People already building developer tools who want their open-source project used by coding agents; maintainers who already have some reputation in a technical community.

Who it doesn't suit: Anyone planning to build, solo, yet another framework for "writing end-to-end tests in natural language". e2e's examples already cover Vite, Next.js, Expo, SwiftUI, Jetpack Compose, Flutter, and .NET MAUI, so there's very little room left if you pick a different front-end framework.

How to do it (if you must build near this space):

1. Don't build a framework. Build a piece it doesn't cover. e2e only has a TypeScript style. It's worth checking first whether the Python (pytest) side has a test library with the same "AI steps + deterministic assertions + cached replay" combination (I infer there's room, but haven't verified). Testing for games, desktop apps, and Chrome extensions is also outside its scope.
2. Take the idea of "cached replay after verification" with you. Any scenario where "letting AI run every time is expensive and unstable" (scraping, automated form filling, data entry) can use the same approach: AI runs it the first time, you record it once it passes verification, and from then on you replay it directly.
3. Design for coding agents from day one: docs bundled into the install package, a one-line init, and it runs directly on a Claude Code or Codex subscription.

How to get users: Copy its launch tweet: a name on the first line, a copy-paste command on the second, five arrows listing selling points below, and one image, all posted at once rather than trickled out as teasers. Clean up the repo before launch: the first screen of the README should be readable code, with a working example for each major framework. After that, post one example repo a day showing "how to solve a specific problem with it", to catch long-tail search. Don't count on Hacker News: e2e itself only got 9 points there.

When to move: The topic "coding agents write the code, who verifies it?" is hot right now, and e2e and Playwright are both rapidly filling in features. To find room next to them, you'd probably have to act in the next few months (estimate).

Signs you read it wrong: The slice you picked (say, a Python version) already has thousands of stars from someone else, or e2e or Playwright officially announces support for it. Or e2e's npm daily downloads drop from 100k back to tens of thousands, showing the "AI testing" heat is fading on its own, so building into it won't pay off.