e2e raises $8M as it shifts from testing agent to open-source framework

The company entered startup accelerator Y Combinator as TesterArmy while Rybczak was still completing secondary school. Earlier backers included the accelerator, Vercel founder Guillermo Rauch and Cognition co-founder Walden Yan.

The founders say daily customer work persuaded them that improving existing testing tools was insufficient. Their response was to build e2e and phase out the earlier product. Notably, the October technical launch post still described the framework as the future foundation of TesterArmy; the funding announcement describes a fuller replacement.

A new framework with existing foundations

End-to-end tests check complete user journeys, such as logging in, buying something and paying. e2e combines natural-language instructions for an AI agent with precise checks written in code. Developers can choose their model rather than being tied to one provider.

The company's October 8 technical launch post, written by Kwaśniewski, identifies an important dependency: browsers run through Playwright, while mobile testing uses agent-device. Its claim to a new framework therefore concerns how testing is organized above those components, rather than replacing every underlying tool.

Nor is agent-assisted testing unique to e2e. Microsoft’s Playwright 1.56 release introduced agent definitions to plan tests, generate test files and repair failing tests in October 2025, a year before e2e’s launch. Those capabilities address building and maintaining a test suite; e2e’s launch post additionally describes model-driven actions within a running test. The descriptions do not establish equivalent capabilities, but they limit any suggestion that established tools lack agent support.

The economic distinction e2e proposes is conditional. Its technical post says the first agent run costs more and takes longer than a scripted step. Successful steps with recorded checks can be cached and replayed without model calls. If interface changes break replay, the live agent returns and model charges resume. That mechanism could reduce repeated inference spending, but the launch post supplies no comparative benchmark establishing the savings across customer workloads.

The same technical post says developers pay their model provider’s token price without an added markup from e2e, and can also use supported subscriptions or local models. For hosted execution, it already supports Kernel’s hosted browsers and Expo’s mobile simulators. That gives users hosting options before e2e launches its own infrastructure. The funding reports supply no pricing for the planned service, and a cached step’s lack of token charges does not establish the total cost of its execution environment.

Company-reported activity since October 1: more than 1 million downloads, more than 12,500 GitHub stars and over 10 million tests. These measures do not identify distinct customers. Source: The Recursive.

What the 2027 launch must establish

In the funding coverage, e2e says that since its October 1 launch it has exceeded one million downloads, 12,500 GitHub stars and 10 million tests. The Czech funding report cites more than eight million tests and 12,000 stars. These are different reported thresholds, without a common measurement cutoff or definitions that would permit a like-for-like comparison. Neither downloads nor test executions identify the number of distinct, continuing customers.

The technical launch post identifies the software’s Apache 2.0 licence, describes it as pre-1.0, warns that parts of the API may change, and limits current platform coverage to web, iOS and Android. Those qualifications matter for teams considering a migration.

The funding will support engineering hires, the open-source team and infrastructure development. The company also says it plans to expand engineers’ direct work on enterprise deployments, with initial projects already underway, according to the account of its deployment plans. The framework is intended to remain free, including for commercial use.

The next decision for users is whether e2e’s planned infrastructure offers enough value to pay for alongside a free framework and existing hosting options. Pricing, measured execution costs, reliable failure detection and evidence of repeat paid use would help answer that question. The financing gives e2e resources to build the service; the reported activity figures do not yet demonstrate its commercial demand.