Medium: Continuous product discovery at scale with AI

What the article is about

Joca Torres published this piece in May 2026 on Medium, addressing a tension that most product teams recognize: continuous discovery is widely understood as valuable, but rarely practiced consistently. Scheduling conflicts, recruitment friction, time pressure, and organizational bureaucracy have historically kept discovery episodic rather than continuous. The article describes how an AI research agent changed that dynamic for teams Torres works with.

Context and method

The agent described is ReveLumi, which conducts qualitative research via WhatsApp. It handles the full research workflow — planning the study, recruiting participants, conducting the interview through a messaging conversation, and producing a report with key themes and suggested next steps. Teams that previously ran user interviews quarterly, when project timelines allowed, have used it to interview ten or more users weekly without adding headcount or changing team structure.

The article positions this as a change in kind, not just speed. Before AI-assisted discovery, the workflow still depended almost entirely on human effort at every step: a researcher planned the study, recruited participants through email or Dovetail panels, scheduled sessions, conducted interviews, took notes, coded themes, and synthesized findings. Each step introduced delay and scheduling risk. ReveLumi collapses this into a coordinated automated sequence where human judgment enters at the planning stage — defining what questions to ask and why — and at the interpretation stage — deciding what the findings mean for the roadmap.

Key takeaways

Torres is careful about what the change does and does not represent. The agent does not replace the researcher’s judgment about what to ask or how to interpret findings against business context. It removes the coordination and execution layer that made research expensive to run. Discovery becomes routine rather than occasional because the marginal cost of another wave of interviews drops substantially.

The article also notes what AI-assisted discovery does not change: the importance of well-formed research questions, the risk of confirmation bias in how findings get framed, and the gap between usability and product-market fit. Torres cites the risk that teams stop sitting with problems long enough to understand them, treating cheap prototyping and fast user feedback as substitutes for harder strategic thinking.

Who it is useful for

Product managers and research leads at teams where continuous discovery is an aspirational practice that has not survived contact with scheduling reality. Also relevant for anyone evaluating AI research tools and wanting a practitioner account of where agent-assisted workflows deliver on their promise and where they require the same judgment as before.