Aakash GuptaThe #1 Skill PMs Need in 2025: AI Product Discovery Masterclass by World’s Leading Authority
CHAPTERS
- 0:00 – 2:10
Discovery in the age of AI: why it matters more when “delivery is free”
Aakash frames the core question: does AI change product discovery or not? Teresa argues that when building/prototyping becomes cheap, discovery becomes even more important to prevent feature bloat, incoherence, and customer fatigue.
- •AI makes delivery/prototyping faster, but doesn’t remove the need to decide what’s worth building
- •Cheap shipping increases the risk of cluttered products and constant-change fatigue
- •Discovery’s purpose shifts from saving engineering time to protecting customer experience
- •Sets up the episode’s dual focus: AI for discovery workflows vs discovery for AI products
- 2:10 – 3:16
Why customer interviews fail: shifting from solution feedback to story-based learning
Teresa explains that many teams interview frequently but still ship failing features because their interviews are unreliable. The fix is to stop seeking validation for solutions and instead collect detailed stories about real past behavior.
- •Most teams pitch solutions in interviews and ask for feedback/approval
- •Hypothetical questions produce optimistic, biased answers
- •Goal of interviews: learn about customers’ real context and behavior
- •Story-based interviewing focuses on what happened last time, not what might happen
- •Improves fit between product decisions and customers’ lived reality
- 3:16 – 5:27
Better questions, better signal: practical story prompts + assumption testing overview
Aakash and Teresa contrast terrible interview prompts (“Would you use this?”) with high-signal prompts (“Tell me about the last time…”). Teresa introduces assumption testing as a separate activity that helps teams learn earlier by testing the riskiest beliefs behind an idea.
- •Avoid asking customers to predict future behavior or evaluate your concept directly
- •Use specific past-event prompts to anchor truthful detail
- •Assumption testing beats ‘big idea testing’ by isolating what must be true
- •Test assumptions faster and cheaper than full prototypes or A/B tests
- •Clarifies division of labor: interviews for understanding, tests for evaluating solutions
- 5:27 – 7:50
Continuous Discovery Habits: outcomes, Opportunity Solution Trees, and compare-and-contrast decisions
Teresa outlines how teams move from output roadmaps to outcome-driven work, then need a repeatable discovery system. She explains the Opportunity Solution Tree (OST) as a way to organize opportunities, generate multiple solutions, and decide via assumption testing.
- •Shift from feature delivery to outcome ownership (retention, churn, engagement)
- •Weekly customer interviews build ongoing understanding
- •OST: outcome → opportunities (needs/pains/desires) → solutions → assumptions/tests
- •Pick a target opportunity, then explore multiple competing solutions
- •Assumption testing supports structured, evidence-based solution selection
- 7:50 – 12:36
AI’s impact on discovery workflows: thought partner, interviewing, and synthesis tradeoffs
Teresa separates AI’s impact into two tracks: helping with day-to-day discovery tasks vs changing discovery when building AI products. She’s optimistic about AI as an aid, but warns against losing empathy and nuance by fully automating interviews or synthesis.
- •AI can help refine outcomes and thinking, but lacks business context by default
- •AI-run customer interviews are possible but risk damaging trust and empathy
- •Synthesis automation: helpful if teams do none, risky if it replaces deep engagement
- •AI summaries can miss 20–40% of nuance that drives true insight
- •Use AI to augment humans; go deeper on differentiating problems/initiatives
- 12:36 – 17:35
AI prototyping and the “feature factory” risk: discovery must provide discipline
Aakash raises the concern that executives can now generate prototypes instantly and demand fast shipping, creating a ‘golden age’ of feature factories. Teresa argues AI prototyping is powerful, but makes discovery and prioritization more critical to maintain coherence.
- •AI prototyping accelerates building—but can amplify stakeholder-driven feature sprawl
- •When delivery is cheap, product coherence becomes the scarce resource
- •“Homer designs a car” analogy: lots of features can still fail the core job-to-be-done
- •Discovery prevents customer exhaustion from constant, irrelevant changes
- •AI prototyping should speed assumption testing, not replace problem discovery
- 17:35 – 21:19
Where AI prototypes belong: assumption-test particulars, not whole ideas
Teresa explains why testing full high-fidelity prototypes is slow, unstructured, and hard to diagnose. She advocates using AI prototypes to test specific assumptions in isolation so teams can pinpoint breakdowns and iterate systematically.
- •Whole-idea prototype tests take time and yield scattered feedback
- •Decompose ideas into steps/assumptions to locate friction precisely
- •Example: building an interview coach required testing multiple workflow assumptions (Zoom/recordings/transcripts)
- •Isolated tests reveal exactly which step fails, especially with third-party constraints
- •Ideal flow: outcome → interviews → opportunity space → target opportunity → multiple solutions → assumption tests/prototypes
- 21:19 – 22:12
Common AI prototyping mistakes: shiny objects without a customer problem
Teresa calls out the most frequent misuse: starting with a cool solution instead of a validated customer need. AI lowers the barrier to “make it real,” so PMs need stronger discipline to anchor work in outcomes and problems.
- •Teams jump straight to solutions with no clear problem framing
- •AI makes it easier to indulge ‘shiny object syndrome’
- •PM job is solving customer needs, not generating clever features
- •Always tie prototypes to a target opportunity and outcome
- •Use discovery to justify why something belongs in the backlog at all
- 22:12 – 24:18
Interviewing skill that changes everything: excavating the story (not guessing)
Even with story-based prompts, many PMs fail by letting interviews become a 50/50 conversation or by leading with guesses. Teresa teaches ‘excavation’—using temporal prompts and deep detail to reconstruct what truly happened.
- •Participants don’t naturally tell detailed, linear stories
- •Interviewers must break the conversational norm and let participants talk more
- •Use temporal prompts: “What happened first? What came next?”
- •Avoid leading/guessing questions that contaminate the story
- •Rich narratives expose friction, decision points, and real-world constraints
- 24:18 – 27:23
Single-interview synthesis: the Interview Snapshot (experience map + opportunities)
Teresa introduces a practical one-page template to turn weekly interviews into actionable learning. The core is an experience map for memory and pattern-finding, plus a list of opportunities (needs/pains/desires) extracted from the story.
- •Separate single-interview synthesis from cross-interview synthesis
- •Experience map captures key moments and preserves recall over time
- •Opportunity list makes interviews actionable for the OST
- •Quick facts add segment/context (customer type, maturity, etc.)
- •Quotes/photos serve as memory anchors; insights capture ‘misc’ learnings
- 27:23 – 28:40
Cross-interview synthesis: building the OST with a “super experience map”
Teresa explains how to move from snapshots to a structured opportunity space. By combining experience maps into a ‘super’ map, teams ensure each OST branch is distinct and can be worked independently with fewer hidden dependencies.
- •Outcome acts as a filter: only move opportunities relevant to the metric into the OST
- •Create a ‘super experience map’ after ~3–4 interviews to represent the full journey
- •Top-level opportunities map to distinct moments in the journey
- •Moment-based structure reduces overlap and dependency between opportunities
- •Supports focused targeting: solve one opportunity at a time
- 28:40 – 30:52
Maintaining a living OST: cadence, iteration, and stakeholder communication
Aakash highlights that the OST is iterative and also a communication tool. Teresa provides a practical cadence: front-load early interviews, draft the opportunity space quickly, then revisit every few weeks as new learning accumulates.
- •Prereqs: clear outcome + 3–4 story interviews + snapshots/experience maps
- •Week 1: front-load interviews to draft opportunity space; Week 2: pick target + start testing
- •Revisit and revise opportunity space every 3–4 interviews
- •Add parent/child structure to organize relationships among opportunities
- •Use the OST to align stakeholders on what’s being pursued and why
- 30:52 – 35:23
Discovery theater and the weekly-interviews debate: what ‘real’ discovery looks like
Teresa lists telltale signs of ‘fake discovery’ and argues the issue is often systemic, not personal. She also softens her stance on weekly interviews, emphasizing agency without shaming PMs whose org context limits best practices.
- •Red flags: backlog never changes, no ideas get killed, teams build the initial idea anyway
- •Failing to consider multiple solutions is a common theater pattern
- •Org incentives often reward certainty over curiosity and learning
- •Weekly interviews are ideal, but being a PM means meeting your company’s expectations
- •Individuals often have more agency than they think, even in constrained contexts
- 35:23 – 48:59
Discovery for AI features: the new product work—context engineering, orchestration, observability, evals
Teresa shares lessons from building her first AI product and proposes a framework for what’s different. AI blurs product/engineering lines and adds new loops: designing reliable prompts/context, orchestrating multi-step systems, logging traces ethically, analyzing errors, and using evals to improve quality over time.
- •Key AI product buckets: context engineering, orchestration, observability, quality (evals), maintenance
- •Prompting in products is ‘one-shot’ and must be reliable at scale
- •Too much context confuses models; right context at right time (RAG/tools/MCP) matters
- •Observability requires trace logging; demands transparency and ethical data practices
- •Error analysis → prompt/orchestration changes → persistent issues become evals for A/B testing
- 48:59 – 52:53
Claude Code for PMs: AI-assisted engineering to ship evals and fixes faster
Teresa describes adopting Claude Code and VS Code to implement an eval, a fix, and an A/B test harness with minimal manual coding. The key is treating the LLM as a powerful but fallible assistant: you must review, understand, and simplify solutions.
- •Claude Code runs in terminal with access to your codebase/files
- •LLM-generated solutions can be overly complex; PMs should guide architecture
- •Teresa shipped a major update + a more advanced eval without writing code herself
- •Best practice: review all generated code and ensure you understand it
- •AI coding tools lower barriers for PMs to participate in AI-product iteration loops
- 52:53 – 56:30
Closing: Teresa’s business, course formats, and wrap-up
Aakash asks about Teresa’s business model and course pricing, revealing a small-cohort, practice-heavy approach designed to change behavior. The episode ends with gratitude, a call for a future round, and the standard subscribe/review outro.
- •Company-of-one model with contractors; focus on impact over scale
- •2,000–3,000 students/year; small cohorts for high support and practice
- •Course tiers: fundamentals ($1,795), deep dives ($799), on-demand ($259)
- •Emphasis on behavior change vs ‘edutainment’
- •Podcast wrap-up, support request, and sign-off
