Silicon Valley GirlAirtable CEO: This Is What the Top 1% Do With AI | Howie Liu
At a glance
WHAT IT’S REALLY ABOUT
How top builders use AI agents to scale judgment fast
- AI has moved from a mature chatbot era into an emerging agent era where models are finally capable of hours of semi-autonomous work.
- The near-term reality is not “companies run by agents while you sleep,” but humans becoming highly leveraged managers of fleets of agents with review and judgment still required.
- Choosing AI tools works best by separating chatbot-style products from frontier agent products, then picking based on autonomy level, integrations, and feedback loops for continuous improvement.
- Practical agent setups can monitor information streams (like X), triage email/Slack, and operate inside team chats (notably Telegram) to learn preferences and deliver feedback at scale.
- The two “superhuman” skills in an agent world are problem-finding (knowing what to automate/build) and judgment (making high-quality decisions across many parallel threads).
IDEAS WORTH REMEMBERING
5 ideasAgents are becoming viable because model capability finally crossed a threshold.
Liu argues “agents were hyped too early,” but newer model generations now support more human-like autonomy, enabling multi-hour tasks and parallel execution.
The winning workflow is human judgment + agent execution, not full automation.
He’s skeptical that agents can run a substantial company end-to-end today; the practical path is increasing the agent-driven portion of the loop while keeping humans as accountable reviewers and decision-makers.
Pick tools by autonomy class before comparing brands.
He recommends first distinguishing chatbot experiences (quick Q&A, limited autonomy) from frontier agents (long-running, tool-using, self-directed work), then selecting based on what your tasks require.
Messaging platforms can become the control plane for teams of agents.
Telegram is highlighted as unusually bot-friendly, making it easy to embed agents in group chats, @mention them for tasks, and let them observe feedback patterns to learn a leader’s “taste.”
A ‘virtual twin’ emerges when an agent has your context and feedback history.
By giving agents access to schedules, communication streams, and prior decisions, users can create an assistant that behaves like them—useful for delegating feedback when the founder becomes the bottleneck.
WORDS WORTH SAVING
5 quotesAnd you could imagine building a company that you never would've dreamed of without any employees or with minimal employees, now you can do that with agents.
— Howie Liu
Like that's the thing I think like a common thread is like, you know, the thing that scales the most and gives you the most leverage in the agent world, and you know, it was true in the human world as well, is good judgment, right?
— Howie Liu
And I think the people who are able to do that most effectively will become like almost superhuman, right?
— Howie Liu
I think what really being a builder comes down to is, one, like having the appetite to tinker. Like, that's the most important thing, right?
— Howie Liu
Problem hunting, like figuring out like what are the problems you wanna solve is like 80% of the battle, right? Because it turns out the actual building of the solution is no longer so hard.
— Howie Liu
High quality AI-generated summary created from speaker-labeled transcript.
