Airtable CEO Howie Liu on 13 years of building, $11B valuation, and the agentic future of databases

Airtable's Howie Liu on 13 years, $11B, and betting on agents

Howie Liu co-founded Airtable in 2013 on a deceptively simple premise: all enterprise software is just a database with app logic and interfaces on top. Thirteen years later, the company has raised over $1B, sits on an $11B valuation from its Series F, holds the full amount on its balance sheet, and is now cash flow positive. The question Liu is focused on now is whether Airtable can remain the data layer as agents increasingly become the interface.

From personal CRM to meta-platform

Liu's path to Airtable ran through a Y Combinator-backed personal CRM called eTacz, which attracted acquisition interest from Salesforce around late 2011. He took the acquihire, spent time inside Salesforce, and came away with a clean architectural observation: Oracle, SAP, and Salesforce are all the same thing underneath — a database with a front end. He wanted to build a simpler, more configurable version that anyone could use, before PLG was even a term.

Airtable spent two and a half years building the product before launching in 2015 — a timeline Liu notes ran almost exactly parallel to Figma's. The first growth assumption was wrong in an instructive way. The plan modeled a long-tail prosumer audience like Dropbox. What actually happened looked more like Slack: the strongest virality came from larger organizations. WeWork, at its peak, had roughly 10,000 employees running operations on Airtable. Liu describes that as early evidence of "data gravity" — once enough data lives in the product, retention and expansion compound on their own.

70-plus percent of Airtable's current enterprise accounts, including those generating $5M-plus in annual revenue, originated from teams inside larger companies adopting the product organically, often as shadow IT when internal IT delivery was too slow.

“Our Series F raised 700 plus million in that round at an $11,000,000,000 billion valuation. And we still have like all of that money on the balance sheet and we're now like cash flow positive. The database layer actually becomes more important with agents — you don't want just a bunch of ephemeral context windows. They need to store and collaborate on data along with humans.”

The enterprise motion

Moving upmarket required swapping viral adoption for consultative selling — coming in to solve a named operational problem rather than waiting for bottom-up spread. Liu frames Airtable's enterprise wedge as analogous to Palantir's: use product flexibility to demonstrate a deep solution quickly, without the distribution scale of a Salesforce or ServiceNow. Some large banks were completely firewalled from the PLG motion and required top-down engagement before any access at all.

AI and the headless question

Liu sees the real AI disruption as two-layered. The surface layer — copilot assistants, AI field agents that can run enrichment or research calls across 20,000 records in parallel — is table stakes. The deeper question is whether users even want to come into Airtable's interface anymore.

His answer is hybrid headless. A pure back-end database (Postgres, Supabase) doesn't solve for permissioning, collaboration, or the human need to occasionally inspect the actual data. The analogy he uses: even if agents generate all your code, you still want to see the diff. Airtable has built a first-class integration with ChatGPT that lets users interact with their data through the chat interface while a fragment of the Airtable view surfaces inline — headless on demand rather than fully decoupled.

HyperAgent

The bigger bet is HyperAgent, a new product built inside Airtable targeting long-running agentic tasks — not ten-second actions but ten-hour autonomous workflows — for non-coders in a business context. Liu frames it as Airtable's version of Amazon's early internet index: just as Bezos picked e-commerce to ride internet growth, HyperAgent is Airtable's way of riding the agent wave directly rather than just adapting to it.

The internal operating model is shifting in the same direction. Liu argues that one agent on one branch can do the work of roughly three engineers at perhaps three times the speed — a 10x leverage factor per agent — and that the best engineers are already overseeing 20 to 30 concurrent agent branches. He expects every knowledge-work role to require the same transition: from individual contributor to manager of a parallel agent fleet.

Liu's clearest conviction is structural. Databases were the right meta-problem to bet on thirteen years ago because they were foundational and durable. Agents, he argues, make the database layer more important, not less — agents need persistent, collaborative, permissioned data stores, not just ephemeral context windows. Whether HyperAgent can claim that position before better-resourced competitors is the open question he is building toward.