Questions
Why did Airtable's CEO say the company's viral 'Airtable is dead' tweet was wrong?
Howie Liu says the tweet's numbers for Airtable's revenue and growth were wrong by a large multiple. He notes that the All-In podcast, which had discussed the tweet, later ran a follow-up saying it had the numbers wrong.
What is an 'IC CEO' and why does Howie Liu say CEOs should become one?
It means a CEO who works as an individual contributor again, building and prototyping things personally. Liu says AI changes so fast that a leader must be in the details to understand what the new tools make possible.
What does 'fast thinking' and 'slow thinking' mean in Airtable's organization?
Borrowing from Daniel Kahneman's book, Liu calls the AI platform group the 'fast thinking' team, which ships new AI capabilities almost weekly. The 'slow thinking' group does more deliberate work, such as the data store that can handle hundreds of millions of records.
Why does Liu think Airtable can beat 'vibe coding' tools at building business apps?
He says an AI that writes every line of code tends to produce bugs, data and security problems, and loses track as the app grows. Airtable's agent assembles ready-made, reliable building blocks instead, and a non-technical user can still see and adjust the result.
When does Liu say you should avoid evals when building an AI feature?
For a brand-new kind of product or capability, he says to start with open-ended 'vibes' testing of different prompts. He says evals, which are repeatable tests of output quality, make more sense once you know the use cases you want it to handle well.
Facts
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
Howie Liu said the author of the viral 'Airtable is dead' tweet appeared to have worked at CB Insights, a company that sells private-company data, yet used incorrect numbers.
“I think they they worked at CB Insights. And the irony is like that the whole point of that business is to have like good data, good data quality around private company data.”
0:05:06 · Howie Liu · Viral tweet
According to Howie Liu, the All-In podcast ran a follow-up episode a few weeks after discussing the tweet, saying it had gotten Airtable's numbers wrong.
“All in did a follow up episode a few few, I think, weeks later saying like, hey, like, you know, we got the numbers wrong.”
0:06:17 · Howie Liu · Podcast correction
Howie Liu said Airtable and Figma were founded around the same time and each spent about two and a half years building their product before launching.
“we both were founded around the same time both spent two and a half years building the product”
0:09:52 · Howie Liu · Company founding
Howie Liu said he is the most expensive user of Airtable AI in inference cost within the company, and for a long time across all Airtable customers as well.
“number one most expensive in inference cost user of airtable ai uh not just within our own company”
0:13:51 · Howie Liu · AI usage
Howie Liu described an Airtable capability, an 'LLM map reduce', that splits large content into chunks, runs a language-model call on each chunk, then runs another call to aggregate the results.
“map through like all of this content and break it up into chunks and then like perform an LLM call on each one and then perform an aggregation LLM call on those chunks”
0:14:48 · Howie Liu · LLM map reduce
Howie Liu said Airtable first reorganized its product and engineering org from feature-based groups into business-unit-style pillars such as enterprise, teams, AI, solutions and infrastructure.
“we reorged initially to basically different business units effectively, right?”
0:20:01 · Howie Liu · Reorganization
Howie Liu said Airtable's 'fast thinking' group is officially called AI Platform and aims to ship new capabilities nearly every week.
“the fast thinking group, which officially is called AI platform. But it really means like we want to just ship a bunch of new capabilities on a near weekly basis.”
0:21:26 · Howie Liu · Org structure
Howie Liu said Airtable's data store, HyperDB, can now handle datasets of several hundred million records.
“our data store, HyperDB, that now can handle like multi-hundred million record data sets”
0:22:27 · Howie Liu · Infrastructure
Howie Liu said Airtable had two layoffs that significantly cut headcount after it grew and overscaled too quickly, though it kept hiring in every major department.
“even when we had to do two riffs, that significantly kind of reduced our headcount, we had just like way too quickly grown and overscaled the business at a certain point”
0:23:51 · Howie Liu · Layoffs
Lenny Rachitsky said ChatGPT's announced figure of about 700 million users refers to weekly active users, roughly 10% of humanity.
“Weekly active users, 10 % of humans on earth use it weekly.”
0:30:41 · Lenny Rachitsky · ChatGPT users · 2025
Howie Liu said that with early GPT-3.5-era models, GitHub Copilot's autocomplete of a few lines of code was the right product form for AI coding.
“the right form factor for leveraging those models in a software creation context was GitHub Copilot. Right. It's like autocomplete a few lines of code at a time.”
0:33:48 · Howie Liu · AI coding
Howie Liu credited Cursor as an early pioneer of an agentic approach to AI coding that generates larger chunks of code.
“I think Cursor did a great job of being an early pioneer of this more agentic way of leveraging the models to do more complex things and generate more larger chunks of code.”
0:34:16 · Howie Liu · AI coding
Howie Liu said Airtable made its AI agent the default way of doing everything in the product, replacing a secondary assistant sidebar.
“We now made our agent the default way of doing everything in Airtable.”
0:32:41 · Howie Liu · Product design
Howie Liu said Airtable's components act like a domain-specific language that its AI agent can assemble to build business apps, instead of writing all the code from scratch.
“the Airtable pieces in our Lego kit today can be used by this agent as almost like a more expressive DSL, like a domain-specific language to build business apps”
0:37:57 · Howie Liu · Agent architecture
Howie Liu said GPT-5 had been released the day before the interview and that he was already testing it on personal use cases.
“GPD 5 came out yesterday, and like playing around with it a bunch, just on like a variety of different, like personal use cases.”
0:42:51 · Howie Liu · GPT-5 release · 2025
Howie Liu cited Runway's release of an immersive world engine as an example of the new AI products he tries out.
“Runway released their like immersive world, you know, kind of engine.”
0:40:38 · Howie Liu · AI products · 2025
Howie Liu said OpenAI's newly available deep research API costs roughly a dollar or more per research call.
“It ends up being like something on the order of like a dollar plus per research call”
1:02:56 · Howie Liu · Deep research API · 2025
Howie Liu said the original product manager specification at Google required PMs to be somewhat technical so they could understand engineering limits.
“the original PM spec at Google required the PMs to actually be somewhat technical so they could understand the engineering”
0:56:18 · Howie Liu · PM roles
Howie Liu said that, as he understands it, a product group at Stripe does not always have the product manager as its directly responsible individual, which is sometimes the engineer.
“on any given product group at Stripe, my understanding is that like, you know, the DRI isn't always the PM, right?”
0:57:06 · Howie Liu · Product leadership
Howie Liu recalled that when building JavaScript single-page apps around 2008 to 2010, rounded corners had to be made as images in Photoshop and positioned by hand.
“if you wanted rounded corners in your ui you literally took photoshop opened it up created like a rounded corner and pixels”
1:28:42 · Howie Liu · Web development · 2008-2010