Lenny's Podcast: Product | Career | Growth · Lenny Rachitsky

How we restructured Airtable’s entire org for AI | Howie Liu (co-founder and CEO)

·1 hr 41 min·5 clips

Howie Liu spends hundreds on AI inference to extract insights from sales calls, calling it trivial compared to the value.

Lenny Rachitsky hosts Howie Liu, co-founder and CEO of Airtable, for a wide-ranging conversation about how an established software company navigates the AI transition. Liu describes confronting the viral tweet and All In podcast coverage claiming Airtable was dead and its valuation underwater, noting the underlying data was factually incorrect and that All In later issued a correction. He then moves to the central theme of the episode: the phenomenon of IC CEOs, CEOs returning to individual contributor work. Liu argues that in the AI era, the details of product decisions are inseparable from the strategic ones, and that you cannot have good taste in a product without actually using the underlying models and primitives yourself. He describes writing more code, calling APIs directly via Postman, and using Airtable AI so aggressively -- including running expensive LLM map-reduce operations across entire sales call transcript corpora -- that he is the highest inference cost user globally across all Airtable customers. Liu explains the rationale for Airtable's recent reorg. Previously structured around feature areas and then business units, the company has now been reorganized into two groups: an AI platform team optimized for speed, shipping impressive new capabilities weekly, and a slow-thinking team handling more deliberate architectural bets like HyperDB, the infrastructure capable of handling multi-hundred million record data sets. He describes the fast-thinking team as requiring entrepreneurial, full-stack thinkers who can hold design, technical, and strategic considerations simultaneously without being overwhelmed by ambiguity. Liu argues every legacy software company should ask itself: if you were founding this company from scratch today, could you execute the mission better with an AI-native approach? If the answer is yes and your existing assets are not net additive, he believes founders should find a buyer and then go start the next iteration. Airtable's advantage, he argues, is that its no-code component building blocks genuinely improve AI-generated business apps by providing more reliable, tested primitives rather than raw code gen, which introduces bugs, security issues, and brittleness. He describes making the AI agent Omni the default primary interface for Airtable, with the traditional Airtable base and table view becoming an artifact that the agent manipulates rather than the primary entry point. He also describes adding code gen capability to Omni so it can extend generated apps with fully bespoke functionality for the final mile of customization. The conversation closes on the product design opportunity: Liu argues AI capabilities are significantly under-merchandised visually, that most AI products present a blank chat box and leave users unable to discover what the system can do, and that building richer visual metaphors and affordances for AI capabilities is a major design opportunity.

As heard by us

A clear look at how an established team is rethinking its mission and workflow around AI.

Howie Liu's conversation keeps coming back to a simple question: if Airtable were starting now, could it be built as a fully AI-native company, or would it make more sense to sell it and begin again in a new form? That gives the discussion real stakes.

Read the full review in PlayNext →

Why you'd press play

You want Howie Liu's take on whether a decade-old product should be rebuilt for the AI era.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Lenny's Podcast: Product | Career | Growth