A founder admits "AI-accelerated research" is just a faster horse, then launches a live in-app interviewer
Building real-time user insights into your design workflow ft. Prayag Narula | Config 2026 · Prayag Narula
19 min total·Actually worth watching closely: ~2 min·1 must-watch clip
- 0:12 – 2:30Listen
Building got faster; understanding users didn't
Opens with the core tension: AI has pushed the speed of building up by orders of magnitude, while the biggest single point of friction — getting real customer feedback — is untouched. The result is that something goes wrong and you only find out after shipping.
The speed gains skipped the "understand your users" step, and turned that weak link into the bottleneck.
The speaker is standing and making an argument; nothing on screen to read. Fine to listen to while doing something else.▶ Jump to 0:12Speaker · Prayag Narula - 2:30 – 5:00Listen
Three feedback routes, each with a fatal flaw
Takes apart why the existing approaches don't hold up: panel respondents may be feeding your questions to an LLM and regurgitating the answer for the incentive; in-app surveys hand users the job of translating their mental model into text, with terrible response rates and fidelity; and interviews simply can't be scheduled.
Teams skip research and go with their gut not out of laziness — every route is blocked.
The three problems are listed out loud, with no diagram or data board. Listening is more efficient than watching.▶ Jump to 2:30Speaker · Prayag Narula - 5:00 – 7:14Listen
The "faster horse" self-critique
Turns the criticism on his own industry: using AI to make interviews bigger and faster is still building a faster horse. The real problem was never that research is slow or doesn't scale.
The insight has to arrive before the decision; once the decision is made, no amount of data collected is worth anything.
The heaviest stretch of argument in the talk, carried by the speaker's wording and pauses. The visuals don't help — worth hearing straight through without skipping.▶ Jump to 5:00Speaker · Prayag Narula - 7:15 – 8:50Watch
Launch: the in-app live interviewer
Announces Marvin's biggest product update to date — an intercept that connects to any website or product and lets AI interview users in the moment they're using it, billed as the world's first.
Picks up the earlier metaphor: this is meant to be the automobile, not a better-fed horse.
Worth a beat on screen when the launch board shows the product name and the claim — the positioning argument that follows keeps referring back to it.▶ Jump to 7:15Speaker · Prayag Narula - 8:50 – 11:00Skim
How the behavior triggers catch people
How it works: JavaScript triggers wait for behavioral signals of frustration and confusion — the same button clicked over and over, a user going back and forth in the navigation, or simply landing on a new app you're designing — and fire an interview on the spot, with the AI recording the screen and asking intelligent follow-ups.
The trigger conditions are a design insight in themselves: frustration has a behavioral fingerprint, so you don't have to wait for a complaint.
The triggers are a short list of parallel rules — if they're on screen, one glance is enough; no need to follow the speaker word by word.▶ Jump to 8:50Speaker · Prayag Narula - 11:00 – 12:58Listen
40 languages and real-time routing back
Rounds out the capability set: the interview asks its questions in 40 different languages, and the resulting insight is pushed in real time into Slack, Teams, Figma — wherever internal stakeholders already work.
How insight gets delivered matters more than the insight itself — it has to appear on the screen where the decision is being made.
A coarse-grained feature rundown; one pass through is enough to retain the point.▶ Jump to 11:00Speaker · Prayag Narula - 12:58 – 16:00Listen
The origin story, and why the moment matters
Where the idea came from, and why "in the moment" matters: interviewing after the fact makes users reconstruct their pathways from memory days, weeks, or months later, while a user sharing their screen and saying "here's why I clicked this" is something of a completely different fidelity for a designer.
The gap between reconstructed memory and live narration is the entire basis for this product.
Narrative and experiential — the speaker is recounting something he lived through, with nothing to see. Good to listen to with your eyes closed.▶ Jump to 12:58Speaker · Prayag Narula - 16:00 – 19:10Listen
One Researcher Per User
Pulls it together into a vision: echoes the conference theme that design and taste are the last true differentiator, insists that even in a world of AI and synthetic users nothing beats talking to real people, and lands on One Researcher Per User — illustrated by the AI interviewer at the booth switching into Spanish on the spot to show what removing every barrier to being heard looks like. Ends with a booth invitation and thanks.
Design can't be a differentiator until we truly understand who we're designing for; the hard part is keeping the human-centered process while moving at the pace of the engineering team.
The value of the close is all in the wording, and the Spanish example is recounted rather than replayed on stage. Listening is enough; the last twenty seconds of thanks can be skipped.▶ Jump to 16:00Speaker · Prayag Narula