Synthetic PM

Competitive research, without the hours

Inspiration

Hi I'm Bo. I build B2B products.

I kept running into the same challenge: it’s difficult to maintain the quality and speed of product thinking as a company grows. Every PM uses LLM, but there’s always more customer feedback to understand, more workflows to test, and more decisions to make.

That led me to a question: Could an AI agent actually do substantial product work—not just write a PRD, but enter a real product, experience it firsthand, and reason about it like a product manager?

This is also part of a larger idea I’m exploring: whether AI can take on complex, non-routine business work that still depends heavily on human judgment—from B2B sales to procurement.

What it does

Synthetic PM explores any online software products, reconstructs its workflows, and turns the experience into a structured product journey.

How we built it

By stitching gpt-5.6-terra, Browserbase, Playwright, human approval gates, structured screen capture, and automated Notion reporting together. All vibecoded via ChatGPT.

Challenges we ran into

Most product feature trade-offs to balance cost and time constraint. Goal is not perfection but to collect enough data to validate the idea through the hackathon. For this product, tech are easier than GTM.

Accomplishments that we're proud of

I am not a coder. Nevertheless, ChatGPT helped me launch in 3 days solo and saved me enough sleeping time. Finding this balance is a true accomplishment for me.

What we learned

I picked Browserbase instead of hiding the agent inside a headless browser because watching the agent work is part of the product. AI can make exploration faster, but it does not change a basic fact that humans are visual animals. Seeing the agent click, hesitate, recover, and discover creates understanding and trust that a final text report alone cannot provide. For our users, they are not just paying for the output, but also the reasoning and experience of interacting with the agent.