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

How a Founder Refounds a Product for the AI-Native Era

In a wide-ranging conversation, Howie Liu, co-founder and CEO of Airtable, sketches a playbook for legacy software companies that must reinvent themselves for generative AI. Liu argues the moment is not merely incremental evolution but a wholesale refounding: products, teams, and leadership mindsets must change. He frames the necessary shift around practical organizational moves, a return to hands-on founder leadership, and a new relationship between human designers and large language models.

From Viral Crises to Hands-On Leadership

Early in the discussion Liu recounts a viral, misleading critique of Airtable and uses it to illustrate how fast narratives spread. That anecdote sets up his larger point: leaders can no longer safely distance themselves from product detail. In the new era, many CEOs are moving back into maker mode — coding, prototyping, and running high-inference experiments to keep pace with rapid model improvements.

Two-Speed Organizations: Fast-Thinking And Slow-Thinking Teams

Liu describes a structural response that has become central to Airtable’s AI strategy: separate teams that operate at different cadences. The "fast-thinking" group ships near-weekly AI-driven experiences, testing agentic features and prototypes. The "slow-thinking" group invests in deep infrastructure and robust platform work that requires careful premeditation. Together they let a company capture viral, experimental value while building durable, enterprise-grade foundations.

Play, Prototype, And Be The Chief Tastemaker

Practical habits matter. Liu recommends leaders and product teams spend deliberate time playing with third-party AI tools, launching weekend projects, and building real prototypes rather than documents. He says empathy for product UX means tasting the soup — interacting with models and primitives directly. That practice produces better visual metaphors, clearer affordances, and more compelling agent experiences than a static spec ever will.

Agentic App Building With No-Code Primitives

Rather than ask an LLM to output full-stack code for complex business apps, Liu favors a hybrid: use agents to assemble reliable no-code building blocks — collaborative data layers, views, automations — as a domain-specific language. This approach combines the speed of agentic generation with the reliability and inspectability of proven primitives, making generated business apps more secure and maintainable.

Roles Converge: Designers, PMs, And Engineers Become Polymaths

AI rewards multidisciplinary skill sets. Liu argues that product managers should prototype, designers should grasp technical constraints and tool-calling, and engineers should think more like product designers. The baseline expectation is no longer narrow specialization but a minimum fluency across the EPD triangle so teams can iterate faster with fewer handoffs.

Measure What Matters: Vibes Then Evals

In very new form factors Liu suggests starting with wide, open-ended "vibe" experimentation to discover what works, then bake in rigorous evals once you’ve narrowed use cases. Evals become the engine of reliable improvement: measure model performance, agentic workflows, and prompt scaffolds to converge on repeatable product behavior.

Across anecdotes and concrete tactics, the conversation centers on a simple thesis: treat AI as a continuous paradigm shift and refound your product around human-plus-agent workflows, fast experimental teams, and founders who remain intimately involved with design and engineering. That alignment — between leadership, speed, and multidisciplinary craftsmanship — is the through-line that turns generative models into durable business value.