AI Agents Use Airtable Primitives as Domain-Specific Language

The AI-Native Transformation Playbook for Established Companies

Howie Liu, CEO of Airtable, shares how he's reinventing a decade-old company for the AI era through structural changes, leadership approach, and product strategy shifts.

The "Fast Thinking vs. Slow Thinking" Organizational Structure

• Split your organization into two complementary groups:

  • Fast Thinking Group: Ships new AI capabilities on a near-weekly basis
    • Focused on rapid experimentation and iteration
    • Creates "jaw-dropping" value with each release
    • Drives top-of-funnel excitement and new use cases
  • Slow Thinking Group: Makes deliberate, premeditated bets
    • Handles complex infrastructure that can't be shipped quickly
    • Enables initial adoption seeds to grow into larger deployments
    • Builds foundations that support scale and enterprise needs

• This structure allows you to:

  • Ship AI capabilities at the pace of AI-native startups
  • Maintain the infrastructure needed for enterprise-grade reliability
  • Balance innovation with stability

The "IC CEO" Leadership Model

• CEOs and leaders need to become individual contributors again in the AI era:

  • Get hands-on with the product and technology
  • Use AI tools hourly, not just daily
  • Be the "chief taste maker" who understands what's possible
  • Reduce standing meetings to focus on timely, urgent topics
  • Lead by example by sharing your own AI experiments

• Why this matters:

  • AI is evolving so rapidly that you can't delegate understanding
  • Every software product needs to be "refounded" for AI
  • The details of implementation matter more than ever
  • You can't taste the soup without participating in creating it

The "Watermelon Picking" Approach to AI Opportunities

• Focus on the abundant, high-value opportunities:

  • "There are so many watermelons on the ground, you don't need to climb the tall coconut tree"
  • Prioritize capabilities that deliver obvious, immediate value
  • Let teams choose which opportunities to pursue rather than prescribing specific features
  • Experiment broadly before narrowing focus

• Implementation strategy:

  • Start with vibes, not evals
  • Test capabilities in an open-ended way before defining metrics
  • Only formalize evaluation frameworks after you've found promising use cases
  • Avoid constraining innovation too early with rigid measurement

The "Play, Don't Work" Learning Method

• Encourage a culture of playful exploration with AI tools:

  • Block out entire days or weeks just to experiment with AI products
  • Create personal side projects to force yourself to use new tools
  • Share your experiments and learning process, not just the results
  • Value curiosity and exploration over task completion

• Practical implementation:

  • Use AI tools in ways unrelated to your core product
  • Try to build something fun that uses multiple AI tools together
  • Spend liberally on inference costs for high-value insights
  • Lead by example by sharing your own AI experiments

The "T-Shaped" Product Team Evolution

• Every role needs to expand beyond traditional boundaries:

  • PMs need to become hybrid PM-prototypers with design sensibilities
  • Designers need technical understanding of what's possible
  • Engineers need product thinking and business understanding
  • Everyone needs minimum competency across all three domains

• Success depends on:

  • Individual attitude toward learning new skills
  • Willingness to cross traditional role boundaries
  • Ability to think full-stack about problems
  • Comfort with ambiguity and open-endedness

The "Clean Slate" Strategic Assessment

• Ask: "If you were founding a new company today with the same mission, how would you execute using a fully AI-native approach?"

  • Evaluate whether your existing assets help or hinder this vision
  • Identify unfair advantages your current product might provide
  • Be willing to dramatically reshape your product and company
  • If you can't leverage your existing assets effectively, consider selling

• For Airtable, their no-code components provide an advantage:

  • Agents can manipulate reliable primitives instead of generating everything from scratch
  • Reduces bugs and security issues compared to pure code generation
  • Provides a more reliable foundation for business applications
  • Allows non-technical users to understand and modify what AI creates