Chapter 20 · NL E2E AI Agents Track No-code

Write the test in plain English, watch an AI run it in a real browser.

BrowserBash turns a markdown bullet list into a live end-to-end journey. Students write the TTA-Cart checkout spec in 9 English bullets; the agent plans them into 22 browser steps, drives the TTACart demo app, verifies every assertion, and writes a Result.md report.

01 What students build

A natural-language E2E suite. The spec is TTA-Cart.md: log in as standard_user, add the "Test.allTheThings() T-Shirt (Red)" to the cart, verify exactly 1 item, check out as Pramod Dutta / 560001, finish, and assert "Thank you for your order!". No selectors, no code, no framework boilerplate: the agent figures out locators, waits and retries on its own, then hands back a verifiable result.

Plain-English Specs

A journey is a markdown bullet list anyone on the team can read, review and edit. The spec is the test: no page objects, no locator maintenance.

Agentic Execution

An LLM planner expands 9 bullets into 22 concrete browser actions, executes them in a live session, and self-corrects when the page needs a wait or a retry.

Verified Results

Every run ends in a Result.md: status, duration, steps executed, and a step-by-step summary of what was verified, with screenshots on failure.

02 Architecture

TTA-Cart.md9 English bullets
──▶
LLM plannerbullets to actions
──▶
Browser agent22 live steps
──▶
TTACart appthetestingacademy.com
──▶
Assertionsverify + retry
──▶
Result.mdstatus · duration · steps
the same live bot gets graded by the chapter 19 deepeval harness · drive it here, judge it there

03 Live demo (simulated with sample data)

This is a UI walkthrough with pre-baked sample data so you can feel a run without a live browser. In the course the agent really drives app.thetestingacademy.com/playwright/ttacart and the numbers below come from an actual passing run.

browserbash · tta-cart.md demo · sample data

04 Inputs & outputs

What goes in

  • A markdown journey spec written in plain English (TTA-Cart.md)
  • Target app URL: the TTACart demo shop
  • Test data inline in the spec: user, password, name, postal code
  • Expected outcomes phrased as "verify ..." bullets

What comes out

  • A live browser run: 22 executed steps for the 9-bullet spec
  • Result.md: status passed, duration 118.5s, steps executed 22
  • A step-by-step summary of every verification, in order
  • On failure: the failing step, the on-page evidence and a screenshot

05 How it's built

Write the journey as a markdown bullet list: actions and "verify" assertions in plain English, test data inline.
The agent parses the spec and plans each bullet into concrete micro-steps: navigate, wait, type, click, assert.
A live browser session executes the plan against the TTACart app, choosing locators from the rendered page, not from a hand-written map.
Each assertion is verified with waits and retries; the "exactly 1 item" and "Thank you for your order!" checks gate the run.
The run is summarized into Result.md with status, duration and executed-step count, ready to attach to a ticket or a PR.
Failed runs keep the evidence: the failing step index, the banner or error text seen on the page, and a screenshot path for the diagnosis.

06 Tech stack

BrowserBash agentPlain-English markdown specsLLM step plannerLive browser sessionTTACart demo appResult.md reportsZero selectors maintained