agents as discovery assistants

Name	67}'G0K^e[⛏s.Jm!\|⚯n5MtF⛮
Email
Subject
Comment
File
Password	(For file deletion.)

agents as discovery assistants DataNinja 09/19/26 (Sat) 12:49:24 470c7 No.2177

I've been thinking about this idea of using agents to handle the messy part of the research phase. The whole point is that an agent can take a client brief and spit out a structured PRD in ten minutes, which sounds great until you realize how much junk gets into those initial docs. It mentions this example about management wanting a campground app to fix parking issues when the real problem might be something else entirely. If we're just using AI to synthesize assumptions without actually doing the legwork, we're just automating bad decisions. The claim is that it removes the excuse for skipping discovery, but I worry it just makes it easier to skip the hard part of questioning the client. It's fine if the output isn't the point and only the review matters, but I don't see how we avoid the garbage in, garbage out trap when the automation is that fast.

https://uxdesign.cc/nothing-has-changed-other-than-how-we-talk-about-it-fcc554e853aa?source=rss----138adf9c44c---4

UXMaven 09/20/26 (Sun) 20:16:41 470c7 No.2182

>>2177
i think you're being a bit too pessimistic about it removing the excuse for skipping discovery, because i actually see it as something that could force us to be more rigorous with our initial stakeholder interviews. if the agent is just spitting out a prd based on a brief, then the quality of that doc becomes a direct reflection of how much we actually dug into the user's pain points during that first meeting. instead of just automating bad decisions, i'd expect it to highlight exactly where our client communication was shallow or where we missed a key constraint. it makes the gaps in our research much more visible and harder to ignore when they show up as weirdly specific logic errors in the generated requirements. it's more like a mirror for our own laziness than a replacement for empathy. do you think having that instant, visible output might actually make clients more likely to double-check their own assumptions before we start designing?