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
Airtable’s AI-native litmus test: rebuild from scratch or sell
Howie lays out a blunt standard for incumbent software companies facing AI: imagine founding the same mission today with an AI-native approach. If you can’t articulate a compelling AI-native execution path—or your legacy product is a net disadvantage—he argues you should consider selling and starting anew.
- •“If we started today” thought experiment as a strategic filter
- •Legacy assets can be an advantage—or technical/organizational baggage
- •AI requires more than bolting features onto an existing product
- •Mission continuity matters more than preserving the current company form
The viral “Airtable is dead” tweet: why it spread and what was actually true
Lenny brings up a viral tweet claiming Airtable was overvalued and “dead.” Howie explains the numbers were wrong by large multiples, why the narrative gained traction, and how amplification by big platforms shaped perception (with later corrections).
- •Incorrect private-company metrics can still go viral
- •Incentives on social platforms reward sensationalism over truth
- •All-In amplification gave the story broader thematic legs
- •Reputational whiplash: corrections rarely travel as far as claims
The rise of the IC CEO: returning to the details as AI reshapes software
Howie describes why he’s spending more time as a hands-on builder again—coding, testing, and shaping product taste. He argues AI is a continuous paradigm shift that demands leaders stay close to UX, models, and technical primitives to keep product-market fit current.
- •Early product-market fit required founder-level technical + UX decisions
- •Scaling pulled leadership away from the craft; AI pulls it back
- •AI evolves rapidly, forcing continuous reinvention of UX patterns
- •CEO as “chief tastemaker” who must actively participate
“Taste the soup” by using AI constantly: tokens, transcripts, and throwing compute at insight
Howie explains that you can’t judge AI products from screenshots; you have to push models and workflows directly. He shares how he uses Airtable AI heavily (even ‘wastefully’) to extract strategic insights from sales calls using expensive but valuable multi-pass LLM workflows.
- •Hands-on usage is required to understand AI capabilities and limits
- •High inference cost can be trivial versus strategic insight value
- •MapReduce-style multi-call workflows for long transcripts and aggregation
- •Treat AI as a high-leverage “chief of staff” for synthesis
How Howie changed his calendar to make building possible again
To create space for deep work and fast iteration, Howie reduced standing 1:1s and shifted toward urgency-driven, insight-based meetings. He favors a ‘barbell’ approach: fewer routine check-ins, more high-quality in-person time, and focused weekly AI execution reviews.
- •Cutting standing 1:1s to avoid schedule lock-in
- •Prioritizing timely meetings driven by real new information (“alpha”)
- •Barbell model: occasional deep relationship time vs constant rituals
- •Weekly sprint cadence for AI execution and competitive pace
Airtable’s org redesign: from feature teams → pillars → fast-thinking vs slow-thinking groups
Howie walks through multiple reorganizations and the trade-offs of each. The newest structure splits execution into a fast-shipping AI platform group and a deliberate slow-thinking group that handles foundational infrastructure and enterprise-grade scaling, designed to complement each other.
- •Feature/surface-area teams can trap organizations in incrementalism
- •Pillar/B.U.-style org improved outcomes but still wasn’t fast enough
- •Fast-thinking AI platform ships near-weekly jaw-dropping capabilities
- •Slow-thinking group builds durable infra (e.g., massive-scale datastore)
What “fast-thinking” requires: autonomous, full-stack product builders (and Omni + code-gen extensions)
Lenny asks what kinds of people thrive in the fast-thinking org. Howie emphasizes autonomy and comfort with ambiguity, then describes upcoming capabilities: Omni can conversationally build apps using Airtable primitives, plus code generation for bespoke “final-mile” visuals and functionality.
- •Successful archetype: entrepreneurial, autonomous, full-stack thinkers
- •Balancing wow-factor UX with model constraints and reliability
- •Omni as conversational app builder using Airtable building blocks
- •Code-gen extensions for bespoke UI elements (e.g., custom map/heatmaps)
New AI form factors: from Copilot autocomplete to agentic app creation
Howie connects model capability leaps to shifts in product UX and form factor. As models improved, the market moved from line-level assistance (Copilot) to agentic IDEs (Cursor) and now to “tell me what to build” app generation—matching Airtable’s original democratization mission.
- •Form factors evolve with model intelligence and context windows
- •Copilot fit earlier models; agentic tools emerged as models improved
- •Cursor as a pioneer of larger-chunk, agentic workflows
- •Airtable’s mission: democratize creation—especially business apps
Airtable’s AI-native strategy: reliable primitives + agent assembly (and a GUI fallback)
Howie argues pure vibe-coding for business apps can be unreliable due to bugs, security risks, and context collapse as codebases grow. Airtable’s advantage is a set of trustworthy no-code primitives (data, UI, automation, collaboration) that an agent can assemble like a DSL—while staying inspectable for non-devs.
- •Business apps need reliability, governance, and maintainability
- •Pure code-gen from scratch risks unreliability and context collapse
- •Airtable primitives act like a DSL/LEGO kit for agents to assemble
- •GUI remains a transparent fallback for iteration and understanding
Making AI experiential: PLG speed, “agent as default UI,” and learning by shipping
Howie explains why AI value must be experienced, not explained in decks—pointing to ChatGPT’s frictionless PLG growth. He describes shifting Airtable to be AI-centric, where the agent becomes the default way to operate and the app becomes an artifact the agent manipulates.
- •AI adoption is strongest when users can try it directly (PLG)
- •ChatGPT’s scale proves the power of frictionless experimentation
- •Airtable is re-centering the whole product around Omni/agents
- •Ship early to learn real use cases; polish later
Empowering teams: play, prototypes over PRDs, and cancelling meetings to explore tools
Howie details how he’s changing expectations across the company: encourage genuine play and curiosity with AI products. He pushes prototypes and interactive demos instead of docs, shares his own AI-generated artifacts openly, and even gives permission to cancel meetings for days/weeks to explore.
- •Play mindset vs box-checking increases learning and creativity
- •Prototypes/demos reveal latency, UX feel, and edge cases quickly
- •Leaders should model behavior by sharing prompts, outputs, and links
- •Explicit permission to block calendars and deeply explore AI tools
Cross-functional skills and collapsing silos: the “full-stack” future for EPD—and beyond
Howie argues AI rewards polymaths more than job titles: designers with technical fluency, PMs who prototype, and engineers who think product. He extends this beyond product teams to marketing and sales, where AI-era effectiveness requires collapsing dependencies and owning outcomes end-to-end.
- •Productivity gains correlate more with attitude than role labels
- •Baseline competency in all three: product, engineering, design
- •PM as prototyper; designers build interactive concepts; engineers lead product
- •Marketing and sales also need “full-stack” ownership (AE + SE fluency)
Evals vs vibes: when to measure, when to explore (double-diamond for AI products)
Howie endorses evals as essential—especially for iteration and improvement—while cautioning against using them too early. For novel AI experiences, he recommends starting with open-ended “vibes” testing to discover the right use-case cluster, then formalizing evals once the scaffold is clear.
- •Evals are powerful after you know what “good” should look like
- •Early-stage discovery needs broad, ad hoc exploration (“vibes”)
- •Use exploration to find the cluster of valuable use cases
- •Then apply evals/AB testing to drive systematic improvements
Counterintuitive startup wisdom: founder mode, holistic thinking, and learning from the “why” behind advice
Howie reflects on scaling pitfalls: industrializing into silos can kill integrative product breakthroughs. He aligns with founder mode as caring about the right details without micromanaging, and shares a meta-lesson: extract the reasoning (“chain of thought”) behind advice, not just the prescription.
- •Factory-style org scaling can erase holistic product bets
- •Founder mode ≠ micromanagement; it’s cross-cutting detail ownership
- •CEOs should stay close to product to avoid incrementalism traps
- •Treat advice like model outputs: the “why” is more transferable than the “do”
Don’t step away from what you love: sustaining the long founder journey + lightning round wrap-up
Howie shares what he’d tell his past self: don’t abandon the craft and details you’re passionate about, even as scale adds operational burden. He closes by encouraging anyone to learn cross-disciplinary skills in the AI era, then ends with quick personal picks, values, and how to reach him.
- •Sustained motivation comes from staying connected to the work you love
- •AI makes learning engineering/design/product more accessible than ever
- •Humility and gratitude as a practical life framework
- •Where to find Howie and how to try Airtable’s AI-first experience
