Vibes-First Discovery Process
by Howie Liu • Co-founder & CEO at Airtable
Howie Liu is the co-founder and CEO of Airtable, a low-code platform for building collaborative apps. Previously, he founded Etacts (acquired by Salesforce) and is known for his deep focus on product design and flexible software architecture.
🎙️ Episode Context
Howie Liu discusses the necessity of 'refounding' a decade-old SaaS company in the AI era, emphasizing the shift of CEOs back to 'Individual Contributor' (IC) roles. He details Airtable's organizational split into 'Fast Thinking' (AI innovation) and 'Slow Thinking' (infrastructure) groups, and explains why product leaders must prioritize 'vibes' and hands-on usage over rigid evaluations in the early stages of AI development.
Problem It Solves
Avoids premature optimization and constraining innovation when building with non-deterministic AI models.
Framework Overview
Instead of starting with rigorous evaluations (evals) or metrics, teams should start with open-ended 'play' to develop an intuition for what the model can do. Formal metrics should only be introduced after the product form factor and use cases have stabilized.
🧠 Framework Structure
Mandatory Play: Encourage teams to bl...
Vibes Over Evals: In the 0-to-1 phase...
Convergent Metrics: Only implement ri...
Prototype over Decks: Don't write PRD...
When to Use
When exploring novel AI capabilities (e.g., long-running research agents) where the 'correct' output is subjective or unknown.
Common Mistakes
Setting up strict success metrics too early, which limits the scope of what the AI is allowed to do and kills serendipitous discovery.
Real World Example
When building an AI crawler, Howie tested it on random queries like 'Find every Marvel movie' to see what happened, rather than defining a success rate metric upfront.
For a completely novel product experience or form factor, you should actually not start with evals and you should start with vibes.
— Howie Liu