Lenny's PodcastWhat AI means for your product strategy | Paul Adams (CPO of Intercom)
CHAPTERS
- 0:00 – 4:28
AI as an existential product-strategy reset: map your core value to what AI can replace or augment
Paul frames AI as a “meteor” that will reshape industries and argues product leaders must revisit first principles: what your product fundamentally does and why customers use it. He outlines a practical approach: map each core workflow to what AI can already do, what it will soon do, and decide where you’ll be replaced vs. where you can be amplified.
- •Start with the product’s core premise and customer problem, not the novelty of AI tech
- •Assess: AI can fully do it, partially do it, or can’t yet—then plan accordingly
- •Replacement vs. augmentation as the key strategic fork
- •Treat AI capability mapping as a roadmap/strategy forcing function
- 4:28 – 7:29
Why failure stories matter: freezing on stage at Cannes and recovering in real time
Paul recounts a keynote at Cannes where he froze, walked offstage while still mic’d, then returned and finished strong. The story becomes a broader lesson about resilience and realizing even “worst case” professional moments are survivable.
- •Over-rehearsing word-for-word triggered a panic/freeze response
- •Walking offstage didn’t end the talk—returning reset his mindset
- •Lasting effect: a small doubt before every talk, but also confidence he can recover
- •Reframing failure as manageable reduces fear and builds adaptability
- 7:29 – 11:01
Google’s failed social bets: fear-driven strategy, privacy blowups, and the Google+ secrecy machine
Paul describes working on Google Buzz and Google+ during the social-tech arms race with Facebook. He argues these efforts were driven by competitive fear rather than customer obsession, leading to missteps like privacy disasters and internal dysfunction from extreme secrecy.
- •Buzz and Google+ were motivated by existential competition, not user need
- •Over-indexing on fear can produce bad leadership and poor product decisions
- •Formative research revealed a real need: better small-group communication
- •Google+’s secrecy (separate building/badges) created internal antagonism
- 11:01 – 12:58
Leaving Google for Facebook mid-project: being treated like a spy and what it taught him about strategy
Paul shares the intensity of departing Google during Google+—including being quarantined and having his laptop forensically analyzed. He connects the drama back to the deeper point: fear-based strategy distorts decision-making and culture.
- •Midstream departure during a confidential initiative amplified suspicion
- •Google’s reaction: quarantine + forensic laptop analysis
- •“Traitor vs. enlightened” perception depends on who you ask
- •Root cause lesson: fear-driven initiatives create unhealthy dynamics
- 12:58 – 15:39
Designing a company, not just a product: Intercom’s ‘ship fast, ship early, ship often’ culture
Paul explains why he joined Intercom: the chance to help shape the company’s culture, not only its interfaces. He details Intercom’s bias toward learning-by-shipping and the inherent tension between speed and craft quality.
- •Owen’s pitch: at Intercom you can “design the company”
- •Embracing big bets implies frequent misses—so the culture must tolerate failure
- •“Ship to Learn” evolved into “Ship fast, ship early, ship often”
- •Managing the constant trade-off: high craft standards vs. rapid iteration
- 15:39 – 17:31
Two camps on AI—and why Paul is all-in: bigger than mobile, possibly bigger than the internet
Paul contrasts AI skeptics (burned by crypto/Web3/metaverse hype cycles) with true believers who see a foundational shift. He places himself firmly in the all-in camp, citing rapid capability leaps like multimodal vision.
- •AI skepticism often comes from recent hype-fatigue (crypto/Web3/metaverse)
- •Paul’s conviction: AI is a step-change on the scale of mobile/internet
- •The “hype → trough of disillusionment → emergence” pattern shapes reactions
- •Multimodal breakthroughs (e.g., vision) signal we’re still early
- 17:31 – 21:25
Making time to learn AI: deliberate reading, paying for tools, and hands-on experimentation
Paul’s advice to busy leaders is blunt: prioritize time for AI or get left behind. He recommends reading continuously, upgrading to access best-in-class models, and actively experimenting to develop real intuition (not secondhand opinions).
- •You don’t need extreme hours—you need priority and dedicated time blocks
- •Reading + trying tools are non-negotiable to stay current
- •Pay for access (e.g., advanced model tiers) to test frontier capabilities
- •Parallel to mobile: leaders who delayed adaptation fell behind
- 21:25 – 23:40
A practical AI strategy method: start with customer value, then audit AI’s capability list
Paul warns against getting distracted by flashy demos and lays out a structured way to evaluate AI impact. He lists concrete AI abilities (write, summarize, search, scan images, take actions) and explains why workflow-heavy SaaS products sit directly in AI’s path.
- •Don’t start with demos—start with the product’s reason-to-exist
- •Create an AI capability inventory (text, images, voice, action-taking agents)
- •Most B2B SaaS workflows are exposed to replacement/augmentation risk
- •Use the mapping to drive clear strategic decisions, not feature bolting
- 23:40 – 26:46
Intercom’s ChatGPT pivot: ripping up the roadmap and building Fin as AI-first support
Paul describes ChatGPT’s launch as a “before/after moment” that forced Intercom to rebuild strategy from scratch. They bet big on Fin—an AI chatbot designed to become the first line of customer support—then expanded it into agent assist within the inbox.
- •ChatGPT triggered a near-total strategy reset from first principles
- •Customer support was explicitly called out as a top disruption target
- •Fin: AI-first chatbot as frontline support, not a human-first flow
- •Fin’s scope expanded: from automation to augmentation inside the inbox
- 26:46 – 29:02
Early impact of Fin: strong deflection rates, but the bigger challenge is organizational change
Paul notes adoption is still early, and impact depends on what you measure (interest vs. financial transformation). Some customers see Fin resolving 50–70% of inbound support, but the harder work is helping teams restructure roles and operations around AI-first support.
- •Success metrics vary: hype/interest is high; revenue transformation takes time
- •Skepticism remains: does it work, and is it ‘as good as a person’?
- •Some customers see 50–70% of questions answered by Fin
- •Biggest hurdle: org design—new roles (e.g., conversation designers) and workflows
- 29:02 – 34:52
Mind-blowing AI capabilities and second-order effects: code, vision, medicine, voice cloning, and translation
Paul walks through the sequence of breakthroughs that changed his mental model: from question-answering to reasoning, coding, and multimodal perception. He then explores job transformation examples (radiology) and societal implications like voice/face replication and real-time translation.
- •From scripted bots to LLM autonomy: far less manual orchestration
- •Code generation raises questions about team ratios and how engineering roles evolve
- •Vision unlocks ‘seeing the world’ use cases—from repair help to translation of signs
- •Medical example: radiology may shift from analysis to verification + patient communication
- •Voice/face replication introduces deep ethical and social implications (e.g., “afterlife” avatars)
- 34:52 – 43:00
How to structure teams for AI: invest in ML depth, but avoid an ‘AI bolt-on’ org model
Paul argues world-class ML engineering is foundational, especially when tailoring frontier models to specific domains like customer support. He explains how Intercom balances specialized ML-heavy projects with product teams that build AI experiences on shared foundations—and why every team should learn AI rather than outsourcing it to a silo.
- •Great ML engineers are the starting constraint for truly differentiated AI products
- •Build on top of OpenAI/Anthropic capabilities; potentially specialize further over time
- •Two modes: ML-heavy foundational work vs. product/design-heavy integration work
- •Avoid the ‘mobile team in London’ mistake: don’t isolate AI to one corner of the org
- •Prefer adaptable generalists plus targeted deep specialists where necessary
- 43:00 – 50:54
Staying current and building conviction amid ambiguity: reading habits, debate, and ‘adapt or die’
Paul shares how Intercom navigates internal disagreement and uncertainty: strong leadership conviction helps, but no one truly knows the end-state. He emphasizes deliberate learning channels (Twitter/X, blogs, newsletters) and making decisions despite high ambiguity—while also actively seeking skeptical viewpoints to avoid bandwagon thinking.
- •Leadership alignment accelerates adoption, but doubt and debate remain healthy
- •Ambiguity is extreme: Horizon 1 plans may be irrelevant in Horizon 2/3
- •Primary learning loops: Twitter/X discovery, company blogs, newsletters (e.g., Matt Rickard)
- •Hands-on testing of tools (e.g., Bard, Rewind) to understand boundaries
- •Read skeptics too; balance optimism with credible counterarguments
- 50:54 – 59:20
Product leadership frameworks: before/after moments, pricing simplicity, and roadmap trade-offs
Paul shifts into pragmatic frameworks he uses to lead: identifying “before/after” moments that demand fresh learning, and approaching pricing as one of the hardest design problems. He shares a core pricing lesson—keep it simple—and introduces a way to evaluate roadmaps via differentiation vs. table stakes.
- •Before/after moments (rebrand, pricing change, tech shifts) require renewed customer learning
- •Pricing is deceptively difficult because value is subjective and multi-dimensional
- •Common trap: compounding add-ons/tiering until customers can’t understand the bill
- •Guiding principle: keep pricing simple; resist complexity creep
- •Differentiation vs. table stakes as a roadmap balancing tool
- 59:20 – 1:12:55
More strategic lenses: pendulum overcorrections, product/market/story fit, and JTBD in practice
Paul explains how organizations repeatedly overcorrect (swinging the pendulum) in product focus and hiring, and why sometimes you must cross boundaries to find the true limit. He then adds ‘story’ to product/market fit, arguing great products fail with confusing narratives, and closes with a grounded take on Jobs To Be Done and the Four Forces for switching behavior.
- •Pendulum swings happen in roadmaps and hiring (experts vs. culture fit; specialists vs. adaptable generalists)
- •Sometimes you must overshoot to discover the real boundary
- •Product/market/story fit: positioning and clarity can make or break adoption
- •JTBD works best when kept simple and tied to research about customer ‘energy’
- •Four Forces helps explain switching decisions (attraction, habits, anxieties, etc.)
- 1:12:55 – 1:23:00
Lightning round: books, mottos, interviewing tactics, and why Guinness tastes better in Ireland
Paul shares personal recommendations and operating principles, including a standout reference-call question that surfaces likely performance feedback early. He closes with favorite products, life mottos, and a fun detour into Guinness freshness logistics and Irish food.
- •Book recs: Paul Arden’s ‘It’s Not How Good You Are…’ and Ray Dalio’s ‘Principles’
- •Reference-call gem: “What feedback will I be giving them in their first performance review?”
- •Mottos: “Only work on what matters most” and “Stop worrying about what you can’t control”
- •Life lesson: being nice goes further than people expect (and can still mean hard decisions)
- •Guinness quality depends on freshness and proximity to brewing; Ireland’s fish is a must-try
