> ## Documentation Index > Fetch the complete documentation index at: https://docs.testbrick.com/llms.txt > Use this file to discover all available pages before exploring further. # Building tests with Bricklayer > How TestBrick's AI builder turns a plain-words brief into a test # Building tests with Bricklayer Bricklayer is the AI that builds your browser tests. You tell it what to test; it drives a real browser, records each step it takes, and hands you a test that TestBrick can replay without any AI. ## Writing a good brief Write the flow the way you'd explain it to a new teammate. One line per step works well. ``` Sign up with a new account. Verify the email. Create a project called "Launch plan". Check the project appears on the dashboard. ``` Tips: * **Say how it ends.** "Check the order confirmation page shows the order number" gives Bricklayer something clear to prove. * **Give logins it needs** — or better, store passwords, PINs and authenticator keys as [Secrets](/ui-testing/secrets) and just name them. * **Leave out the email.** If a flow needs an email address, Bricklayer uses a TestBrick inbox so it can read the code or link that arrives. See [Email & codes](/ui-testing/email-otp). * **One journey per test.** "Sign up" and "change password" are two tests. A brief for Bricklayer The brief box shows roughly how many [AI credits](/account/ai-credits) a build uses. ## While it builds Bricklayer building * **The live browser** on the left shows what Bricklayer is doing right now. * **The chat** on the right lists each step it saved. Click the eye to see the page at that step; click the bin to drop a step you don't want. * **Type in the chat** to steer it — "use the second product instead", "the button is under the menu". It reads your message before its next move. * **When it's stuck**, it says exactly what's confusing it and asks you. * **Take control** lets you do a tricky part yourself in the live browser — a hard-to-reach menu, a drag-and-drop. Your steps are saved too, marked "you did this". Hand back when you're done. * **Stop & save** ends the build early and keeps the steps so far. * **Hide** tucks the build away; it keeps going and you can come back to it. Bricklayer never pays, sends money or deletes things on its own initiative beyond what your brief asks for. ## Confirming the checks When it's done, Bricklayer shows how it will prove the flow still works on every future run — usually where the flow ends and what the final page shows. Keep the checks you agree with. It also lists the values it typed. Switch a value to **unique each run** when your app only accepts it once (a new email, username or phone number on a sign-up). Leave logins exact — the same account is the point there. Click **Save & verify my test**. ## The first replay TestBrick then replays the new test once, straight away, without Bricklayer. This proves the test runs on its own. The first replay * **Green** — your test is saved and ready. * **A check failed** — you see which one and why, and can fix it before relying on the test. ## Stages Bricklayer also saves the **stages** of your flow — for example "log in", "add to cart", "open the cart" — with how to tell each one is done. They appear as headers above the steps. Steps grouped by stage Stages help later runs stay honest: * If your app changes and a stage is already done (say, a welcome screen was removed), its leftover steps are skipped instead of failing. * If a run can't get past a stage, the result says "stuck at stage 3 of 5" instead of a vague error. * If a run finds a new way through a stage, it offers to **save the new way** — you decide. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.