Lenny's PodcastHowie Liu: How Airtable refounded its product for the AI era
Through fast and slow-thinking org splits and an IC-CEO who cut one-on-ones; Liu still ranks as the number-one inference-cost user of Airtable AI.
CHAPTERS
- 0:00 – 4:28
Cold open: The AI-native litmus test—refound or sell
Howie opens with a blunt framework for legacy software companies in the AI era: imagine founding the company today with the same mission, but fully AI-native. If you can’t credibly answer that, he argues you should consider selling and starting the “next incarnation.” This sets the episode’s core theme: AI isn’t an add-on; it forces a reset in strategy and execution.
- •Use a “found it today” thought experiment to test AI relevance
- •If the AI-native version isn’t better with your current assets, consider exiting
- •Treat AI as a re-founding moment, not incremental feature work
- •Mission continuity matters more than preserving the existing org/product shape
- 4:28 – 8:10
The viral “Airtable is dead” tweet—what actually happened
Lenny asks about a viral tweet claiming Airtable was “dead,” overfunded, and underwater. Howie explains the numbers were wrong by large multiples and reflects on why sensational, incorrect data spreads faster than corrections. He also notes how amplification (e.g., big podcasts) can turn a bad take into a broader narrative.
- •The tweet’s underlying metrics (revenue/growth) were materially incorrect
- •Virality was driven by sensational incentives on social media
- •All-In later issued a correction, but the meme had already spread
- •Lesson: narratives often outpace truth and are hard to unwind once viral
- 8:10 – 12:14
The rise of the IC CEO: why leaders must get back in the weeds
Howie describes a shift back toward hands-on building—coding, prototyping, and direct product engagement—especially for software-first businesses. He argues that for products like Airtable, “the tech is the product,” and you can’t separate strategy from the detailed design and architectural choices. In AI, this becomes even more essential because new model capabilities constantly demand new UX patterns.
- •Early PMF required founders to make intimate technical and UX decisions
- •Scaling pushes CEOs toward process/people, away from product details
- •AI changes faster than prior platform shifts (mobile/cloud), requiring constant reinvention
- •Being the “chief tastemaker” requires participating in making the product
- 12:14 – 16:27
“Taste the soup”: daily AI usage, prototypes, and compute as leverage
Howie explains why executives need direct, frequent interaction with AI systems—not just reading about them. He shares how he uses Airtable AI heavily (even “wastefully” by traditional standards) because the strategic payoff dwarfs inference costs. The bigger message: aggressively apply compute to high-value questions and let real outputs shape decisions.
- •You can’t understand AI capabilities without hands-on experimentation
- •Howie tracks (and leads) inference usage as a proxy for immersion
- •High-cost workflows (e.g., MapReduce-style LLM processing) can be trivial vs. strategic value
- •AI can replicate “consulting-grade” synthesis from large corpora (e.g., sales calls)
- 16:27 – 18:55
Making time for building: fewer standing 1:1s, more urgency-driven meetings
Lenny presses on what “IC CEO” looks like in practice. Howie explains he reduced recurring one-on-ones to protect focus for timely, insight-driven work. He prefers a barbell approach: fewer forced rituals, more high-quality relationship time in person, plus meetings anchored on fresh “alpha” and urgent product decisions.
- •Cut standing 1:1s to reduce calendar fragmentation
- •Prioritize meetings triggered by new insights and timely decisions
- •Use in-person time for deeper, less-structured relationship building
- •Create a cadence for AI execution reviews to maintain intensity and momentum
- 18:55 – 23:32
Airtable’s AI-driven reorg: from feature teams to fast-thinking vs. slow-thinking groups
Howie walks through Airtable’s org evolution and why prior structures still felt too slow for AI. The latest reorg splits the EPD org into “fast thinking” (AI Platform) shipping jaw-dropping capabilities weekly, and “slow thinking” teams handling deliberate, infrastructure-heavy bets. The two modes complement each other: fast creates excitement and new use cases; slow turns adoption into durable scale.
- •Feature/surface-area teams tend to produce incremental thinking
- •Business-unit structures improved holism but didn’t match AI-native shipping speed
- •Fast-thinking group: rapid, high-impact AI capability shipping on near-weekly cadence
- •Slow-thinking group: deliberate infrastructure/scale work (e.g., databases) that can’t be rushed
- •Together: fast drives top-of-funnel; slow enables retention/enterprise expansion
- 23:32 – 32:58
What great “fast thinking” builders look like: autonomy, full-stack product sense, and open-ended design
Lenny asks who thrives in the fast-thinking team. Howie emphasizes entrepreneurial autonomy more than pedigrees, and values people who can blend product taste, design judgment, and technical constraints. He illustrates with Omni: conversational app building plus code generation for bespoke “final-mile” visuals and behaviors—work that’s inherently ambiguous and demands cross-disciplinary thinking.
- •Successful fast-team members operate autonomously and entrepreneurially
- •The key is full-stack thinking (UX + technical constraints + wow factor)
- •AI product work is open-ended; candidates must enjoy ambiguity
- •Example direction: Omni app generation plus bespoke code-gen extensions for custom UI/visuals
- 32:58 – 46:33
New AI form factors: why “tell me what you want to build” is just the beginning
They discuss the shift toward prompt-first app building and why it emerged as models improved. Howie connects model capability to product UX evolution: early models favored Copilot-style autocomplete; stronger models enabled agentic IDEs and full app generation. Airtable’s thesis is to align its original mission—democratizing business app creation—with AI-native workflows while keeping reliability and user control.
- •Model improvements drive new UX patterns (autocomplete → agentic building → full app generation)
- •“Vibe coding” is magical, but reliability and maintainability are issues for business apps
- •Airtable focuses on business apps (CRM, inventory, case management), not consumer games
- •Airtable’s primitives act like a DSL/LEGO kit agents can assemble reliably
- •Fallback to GUI keeps non-technical users unblocked vs. opaque code-only systems
- 46:33 – 50:56
Company-wide expectations in the AI era: play, prototypes over decks, and faster iteration loops
Howie describes how he’s reset expectations for teams: “play” with AI tools in a real exploratory mindset, not just box-checking. He encourages people to cancel meetings to experiment, and he leads by sharing his own AI-made artifacts (landing pages, research outputs, prompt trails). Execution shifts toward interactive prototypes and open-ended testing rather than polished PRDs and slideware.
- •Encourage true play/curiosity to learn faster and discover new affordances
- •Give explicit permission to block a day or week to explore AI tools
- •Model behavior by sharing prototypes, links, and prompt traces internally
- •Prefer functional prototypes/demos to decks and deterministic timelines
- •Test non-golden-path scenarios early to understand speed, UX, and failure modes
- 50:56 – 1:03:36
Who benefits most from AI tools—and why roles must become ‘full-stack’
Lenny asks which function (PM/Eng/Design) gains the most from AI. Howie argues it’s less about role and more about mindset and polymath ability: designers who grasp model/tooling constraints, engineers with product taste, and PMs who prototype instead of only writing docs. He predicts every role needs a baseline competency in the other two, and extends the same logic beyond product to marketing and sales.
- •Impact depends more on attitude than job title
- •Hybrid skill sets (“polymaths”) gain the biggest leverage from AI tools
- •PMs should evolve into PM-prototyper hybrids with design sensibilities
- •Design becomes more interactive and systems-oriented; engineering benefits from product fluency
- •Role-collapsing extends to marketing (copy + ops + creative) and sales (AE + SE fluency)
- 1:03:36 – 1:12:41
Evals vs. vibes: how to test AI products without constraining discovery too early
They dig into evals as a critical AI product skill, with Howie adding a nuance: don’t start with rigid evals when the form factor is new. Begin with “vibes”—broad, open-ended exploration to discover what works and what users will actually do. Once use cases and product scaffolding are clearer, then formalize evals to iterate and improve systematically.
- •Evals are essential—but premature evals can lock you into the wrong framing
- •Start with open-ended ‘vibes’ testing for new form factors
- •Use exploratory prompting to find clusters of valuable use cases
- •After convergence, codify evals/benchmarks to measure improvements and regressions
- •Merchandising may evolve from open-ended capability to more guided, structured workflows
- 1:12:41 – 1:25:44
Counterintuitive founder wisdom: avoid over-industrializing and don’t step away from what you love
Howie reflects on scaling lessons that echo ‘founder mode’: specialization and fiefdoms can increase local efficiency but kill holistic breakthroughs. He argues CEOs must remain deeply product-engaged to drive step-function innovation—especially when markets shift. If he could advise his past self, it’s to stay close to the product details he loves, even as operational burdens grow.
- •Scaling advice often pushes leaders toward delegation and disconnected swim lanes
- •Over-compartmentalization sacrifices integrative thinking and step-change product bets
- •Founder mode is not micromanagement; it’s caring about cross-cutting details that matter
- •Meta-lesson: interrogate the ‘why’ behind advice, not just the prescription
- •Personal advice: don’t abandon the product/UX craft that energized the company’s origin
- 1:25:44 – 1:40:41
Closing: learnability, the best time to become a builder, and lightning round
Howie ends with a call to action: AI-era relevance is learnable, not innate, and everyone can become more multi-disciplinary through practice and projects. He highlights how AI tutors and tools dramatically lower the barrier to building and learning. The lightning round covers books, entertainment, favorite products, life motto (humility and gratitude), and where to reach him.
- •AI-native skill growth is accessible through practice, projects, and experimentation
- •Modern AI tools act as always-on tutors for architecture, coding, and product thinking
- •Building software is far less arcane than past eras (fewer barriers to creation)
- •Lightning round: books (Three-Body Problem), TV (The Studio), product picks (Runway; Japanese apparel)
- •Life framework: humility + gratitude; contact: HowieTL and howie@airtable.com
