Week of August 8–14, 2026
UI Workflows 🔥
Until now, Test Suites could only test APIs directly. Now you can test through the browser: create a UI Workflow suite and describe each step in plain language: "fill in email and password, then sign in," "click the Request Access button." No selectors, no CSS, no XPath. Each step is resolved against the live page by an LLM reading a snapshot of what's actually on screen, and executed action by action until the step's instruction is fully carried out.
Checks work the same way. Alongside the deterministic checks you'd expect (element present, element absent, URL matches), UI Workflows add a vision-based qualitative check. Based on the step's screenshot, you can describe what should be true ("the confirmation banner is visible and mentions the order number"). It's evaluated the same way a text response would be, screenshot included.
Objective tab and Intent Evaluation 🔥
Every Test Suite now starts with an Objective tab, where you set the suite's type (API or UI Workflow, locked in at creation) and, optionally, its Intent: the goal a real user should be able to accomplish. Once an Intent is set, it changes how runs are judged: instead of a flat percentage of checks passing, an evaluator reviews the Intent alongside every step's results and makes a holistic pass/fail call. A suite can pass with a failed check if the underlying goal was still met, or fail despite every check passing if it wasn't.
Results export
Export a run's full results to Excel. Every step, check, and extracted value, with screenshots included for UI Workflow runs. Used for sharing outside the app or archiving alongside a release.
In the Lab
New this week: When AI Writes the Code and the Tests, What Does Passing Actually Mean? It looks at a July 2026 benchmark (RepoRescue) where AI coding agents "fixing" broken repos leaned heavily on rewriting the tests rather than the code. It poses the question, what's a green test suite actually worth when the same system that wrote the code also gets to grade it?