Use a browser agent, such as the one Ezra calls Astra in Codex to systematically test software changes based on a documented test plan. The agent diligently follows instructions to catch edge cases and inconsistencies that a busy human might miss, writing its findings back to a shared document for review.
with Claire Vo
In Codex, I have a little bug bash. I have a threaded conversation that just lives here all the time. And when something new comes up for our team... we have these FigJam files that explain, 'Here's how to bug bash it. Here's the test account to take, here's the path to follow'... I have started basically snagging one account for me and one account for Astra and just having it run through... it will run through all of these scenarios, and it will use the browser, and it's a dream. And then…
A FigJam board populated with comments from the AI agent detailing the bugs and issues it discovered during the test run.
OpenAI's cloud-based AI software engineering agent that can execute code, run tests, and handle complex multi-file tasks autonomously.
Figma’s collaborative online whiteboard for planning and workshops.
Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.
5 steps
In a FigJam file, document the exact steps for testing a feature. This plan is the primary input for the agent.
Be explicit. Include which test account to use, the URL to start at, the exact sequence of clicks and inputs, and what the expected outcome of each action is. The more detailed the plan, the more effective the agent will be.
Open a persistent conversation thread in your browser agent (e.g., Codex). This thread will be dedicated to the bug bash.
This is an example prompt: "You are Astra, my QA partner. We are going to bug bash the new seat request modal. Your test account is [test_account_email]. Follow the test plan located in this FigJam file: [link_to_figjam]. Be diligent and execute every scenario."
Assigning the agent a specific persona and test account helps it maintain context throughout the session.
Instruct the agent to begin working through the scenarios in the test plan. The agent will take control of a browser window to perform the actions.
This is an example prompt: "Please begin with scenario 1 in the test plan. Document any deviations from the expected behavior, errors, or visual glitches you encounter. When you are finished with all scenarios, let me know."
The key instruction is for the agent to be more diligent than a human and follow the directions precisely, rather than taking shortcuts.
Once the agent has completed the test scenarios, instruct it to add its findings as comments on the original FigJam board.
This step centralizes all findings—both human and AI-generated—in one place, making triage easier for the product manager.
As the product manager, review the comments left by the agent on the FigJam board. Evaluate the severity of each issue and create tickets for the engineering team as needed.
The human remains the ultimate decision-maker on what constitutes a launch-blocking bug versus a minor nit to be fixed later.
Turn an idea into a PRD, user stories, and a plan.
How to recover when the loop fails and where human judgment helps.
Let routine work keep moving.
Create a repository-based system to provide an AI assistant with durable context about your team, projects, and communication channels. This enables it to perform routine tasks like summarizing daily updates and identifying stakeholders, acting as a personalized chief of staff.
Use Meta's Muse to generate a printable, multi-page family newspaper that includes a weekly schedule, weather forecasts, and activity ideas. This workflow leverages Muse's strength in creating visually appealing, family-oriented artifacts.
Coordinate agents in Slack and Cursor to propose growth experiments for human review. Ezra’s Auto XP system was still being developed at the time of the episode.
Join 250,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.
