Will Zhao

Co-founder & CEO @ Shiplight AI. Ex-Airbnb, Snap, Facebook. Entrepreneur, investor, learner.

[Will Zhao @nerdwillz]

$229B in AI revenue. Stocks moved only when a disclosed revenue line appeared. Even Wall Street pays on receipts now.

[Will Zhao @nerdwillz]

@ScendInc @ShiplightAI Yes! Check it out

[Will Zhao @nerdwillz]

E2E tests break for two reasons: the product broke, or the test went stale.

Most frameworks can't tell the difference. So teams either burn hours triaging or start skipping tests.

We built @ShiplightAI to tell the difference. It's open to everyone today.

[Will Zhao @nerdwillz]

@OmMistry2502 @ShiplightAI Thanks!

[Will Zhao @nerdwillz]

@schkolnick__ @ShiplightAI Thank you Matt!

[Will Zhao @nerdwillz]

@shiv_kampani @ShiplightAI Thank you Shiv!

[Will Zhao @nerdwillz]

@ShravanGReddy Thanks! @ShravanGReddy

[Shravan Reddy @ShravanGReddy]

Coding agents can now ship faster than teams can verify the results of the work. That makes independent verification more important than ever.

Shiplight is building the layer for verification: tests that capture product intent without indexing on brittle implementation details. Congrats to @nerdwillz & Feng on opening Shiplight to everyone!

[Will Zhao @nerdwillz]

E2E tests break for two reasons: the product broke, or the test went stale.

Most frameworks can't tell the difference. So teams either burn hours triaging or start skipping tests.

We built @ShiplightAI to tell the difference. It's open to everyone today.

[Will Zhao @nerdwillz]

HeyGen's head of QA went from 60% of the week authoring and maintaining tests to 0% in a month.

Full feature set on the Free plan. One line to install for Claude Code, Cursor, Codex, and more.

Start testing free → https://t.co/V2cQLyxCZM

Ship faster. Break nothing.

[Will Zhao @nerdwillz]

At run time the YAML transpiles to Playwright + Agentic SDK, so everything Playwright supports works and you can eject whenever you want. No proprietary runtime holding your suite.

Resolved locators are cached, so a passing step re-runs with no model call.

[Will Zhao @nerdwillz]

When a test drifts, /shiplight fix reproduces the failure, finds the root cause, repairs the YAML, re-runs it to verify, and opens a PR.

When the app is what broke, it files the bug instead of "fixing" the test to pass. That distinction is the product.

[Will Zhao @nerdwillz]

A Shiplight test is readable YAML in your repo:

goal: user can complete checkout
steps:
- intent: add the $40 item to the cart
- intent: pay with the saved card
- VERIFY: confirmation shows the right total

Your agent writes it by walking your app. You review it like a spec.

[Will Zhao @nerdwillz]

@VictorTaelin @morew4rd Have it open the browser and see how bad it is for itself haha

[Will Zhao @nerdwillz]

@tom_doerr Cool stuff!

[Will Zhao @nerdwillz]

@garrytan It's alright at making tests, there are ways to give it scaffolding to make those tests completely e2e

[Will Zhao @nerdwillz]

@DeRonin_ I'll add one more for you - if you want end to end testing on what you did make, add the Shiplight plugin and get easy, full coverage
https://t.co/o22GVCKhCC

[Will Zhao @nerdwillz]

@SirDanJets @ycombinator @ArgaLabs Good question

[Will Zhao @nerdwillz]

AI made writing checks cheap.

it did not make senior engineering time cheap.

the real cost shows up when experienced engineers lose mornings inside triage loops that should already have a clear path to resolution

[Will Zhao @nerdwillz]

the hidden org problem in AI-generated browser testing:

nobody explicitly owns the ambiguity loop

the failure happens
everyone sees red
and the team loses time deciding whether it’s app, test, infra, or config

the slowdown isn’t the issue itself

it’s unclear ownership of the path to resolution