What Is Airtable? Complete History: Howie Liu, Superagent, Hyperagent, Omni & the AI Refound (2026)

Airtable is the company that turned a spreadsheet into a database that builds apps. In 14 years it went from "you'll never get developers to use a no-code tool" to a $11.7 billion decacorn powering more than 500,000 organizations, then through two layoff rounds and a viral "Airtable is dead" tweet, and finally to a CEO-led AI refound that ships near-weekly capability releases plus two brand-new agent products on their own domains.

This is the complete history of Airtable — the founders, the funding, the product layers, the people, the layoffs, the refound, Cobuilder, Omni, the DeepSky acquisition, Superagent, and Hyperagent. 🔮

TL;DR — From spreadsheet to Superagent in 18 months. Airtable went from a 2012 spreadsheet-meets-database side bet to a $11.7B Series F decacorn, then through 491 layoffs and the "Airtable is dead" meme cycle. Between June 2025 and February 2026 it rebuilt itself as an AI-native app platform: Omni (June 24 2025 conversational refound), the DeepSky acquisition (October 2025, former OpenAI engineering leader for ChatGPT Business Products), Superagent.com (January 27 2026 — first standalone product in 13 years), and Hyperagent.com (early 2026 — cloud agents with their own compute). Build the same shape — but with embedded agents and one-prompt deploy — in Taskade Genesis →

🤖 What Is Airtable?

Airtable is a low-code platform that combines a relational database, a no-code interface designer, and workflow automations into one workspace. Founded in 2012 by Howie Liu, Andrew Ofstad, and Emmett Nicholas, it serves more than 500,000 organizations and roughly 80 percent of the Fortune 100 — including OpenAI, Anthropic, Amazon, Netflix, Google, and Airbnb. As of 2026 the company reports approximately half a billion in annualized revenue with $100M+ in free cash flow.

The mental model is three layers stacked on top of each other:

┌────────────────────────────────────────────────┐
│  INTERFACES   ← what end users actually touch  │
├────────────────────────────────────────────────┤
│  AUTOMATIONS  ← background workflow logic      │
├────────────────────────────────────────────────┤
│  DATA         ← tables, fields, relations      │
└────────────────────────────────────────────────┘

Behind the familiar grid sits a typed, relational schema. Cells are typed: a date field rejects text, a single-select field is constrained to defined options, a link-to-another-record field threads bidirectional relationships between tables. Above the data sits a logic layer of triggers and actions; above that sits a designer that turns the database into something a non-builder can click through.

That three-layer separation is the wedge. Excel and Google Sheets are flexible but unstructured; Salesforce is structured but inaccessible. Airtable's bet was that the middle of the spectrum — typed, relational, but consumer-grade — was an entire category nobody had built.

📊 Airtable at a Glance: The Complete Evolution

The 14-year arc, from Howie Liu selling Etacts to Salesforce in 2010 through the Hyperagent public open in April 2026:

Snapshot table

Snapshot as of May 2026. Funding and valuation figures cross-checked against Airtable's own newsroom posts, TechCrunch, Bloomberg, and CNBC. Revenue is Liu's own framing on the Greg Eisenberg podcast (Apr 2026).

🥚 Founders & Founding (2010–2014)

Airtable was founded in 2012 by Howie Liu, Andrew Ofstad, and Emmett Nicholas — three engineers who met through Duke connections and wanted to consumerize the relational database. The seed of the company sat inside Salesforce, where Liu had landed two years earlier after the Etacts acquihire.

Howie Liu's path: Duke → Etacts → Salesforce → Airtable

Born in 1989 in College Station, Texas, Howie Liu taught himself C++ at age 13 and entered Duke University at 16. He graduated in 2009 with degrees in Mechanical Engineering and Public Policy. In February 2010, at age 20, he co-founded Etacts with Evan Beard through Y Combinator's W2010 batch — a Gmail and IMAP relationship-management tool that pulled inbox metadata into a usable CRM surface. Etacts raised about $650,000 from Ron Conway, Eric Hahn, Jim Young, Lorenzo Thione, Barney Pell, Joshua Schachter, Jawed Karim, and Ashton Kutcher.

On December 21, 2010, Salesforce acquired Etacts for an undisclosed sum. Etacts shut down on January 31, 2011 and Liu joined Salesforce as a product manager working on social CRM. Force.com — Salesforce's developer platform — was scaling fast around him, and the question that became Airtable started forming: what would Force.com look like if a non-developer could use it?

Andrew Ofstad and Emmett Nicholas

Andrew Ofstad grew up in rural Montana, studied Electrical Engineering and Economics at Duke, and spent years at Google as a Product Manager on Android before leading the redesign of Google Maps. Emmett Nicholas was a Stack Overflow software engineer for three-plus years before joining the founding team. The three connected through Duke; the often-repeated "all three were classmates" line oversimplifies — they overlapped but in different cohorts.

How the founders connect: Force.com → Etacts → Airtable → Hyperagent

The throughline of Howie Liu's 17-year crusade reads as one decision tree:

Why investors said no

Airtable raised pre-seed angel checks from Ashton Kutcher, Michael Birch (Bebo), and Josh Reeves (Gusto) in 2012 and 2013, but the institutional VCs were skeptical. Liu summarized it in his 2022 Upfront Summit conversation with Axios:

"Every investor at the time thought we were nuts. They said, 'You can build a great product, but people will not come.'"

The thesis was hard to articulate before "low-code" was a category and before the consumerization-of-enterprise wave (Miro, Figma, Zoom, Slack) had crested. Airtable spent two years in stealth and then opened an invite-only beta in 2014.

A note on the founding-year ambiguity: Wikipedia, Crunchbase, and Tracxn cite 2012. Contrary Research cites January 2013. Golden encyclopedia cites September 2012. The 2012 framing is most-cited and aligns with the founders' own retrospectives — but some company materials use 2013 as a formal incorporation date.

The patient three-year incubation: how Airtable was different from Etacts

The founding decisions Liu made in 2012 were a deliberate inversion of the ones he made in 2010 with Etacts. He told the Startup Grind audience in 2018, right after the Series B, that he had finally given himself the luxury Etacts never had: time. That difference shows up in eight specific design choices that compounded into the eventual decacorn.

Liu's framing of the Excel positioning is worth quoting directly because the lesson generalizes to every horizontal product:

"We're not trying to be categorically better than Excel, but rather, you know, for this large class of use cases we can be much better — but for the numerical use cases we're not the right tool, you should use Excel."

— Howie Liu, Startup Grind interview, 2018

The invisible product decision baked into that framing: Airtable shipped templates on launch day specifically to make the use-case wedge legible to first-time visitors who couldn't otherwise tell what a "consumer Force.com" was for. Today every horizontal AI product faces the same cold-start problem; templates and gallery patterns are the analogue.

The Etacts vision Liu still kept: an "automated and intelligent" CRM

The Etacts pitch in 2010 — described by Liu on Startup Grind as a CRM "that would automatically pull in your emails and your phone calls, Facebook, Twitter, LinkedIn, you know, kind of graph, and show you an evergreen dashboard of all the people you had or hadn't talked to" — reads in 2026 like a description of what AI-augmented relationship-management products are now finally shipping. Liu's mistake with Etacts was timing, not vision. The same automated-and-intelligent instinct survived through Airtable's three product layers and is the throughline that makes Omni and Hyperagent feel like the natural endpoint of a 17-year crusade rather than a strategic pivot.

The Ashton Kutcher meeting (literally) on the set of Two and a Half Men

The most-told Airtable origin anecdote is also true: Liu pitched Ashton Kutcher for Airtable in his trailer in Burbank, right after Kutcher walked off the set of Two and a Half Men. Kutcher had previously invested in Etacts as part of his first year attending YC demo days, so the relationship existed, but the geography was real. Liu's framing on Startup Grind is matter-of-fact: Kutcher "is an extremely savvy, extremely sharp tech investor that holds his own or better against any of the other top tech investors." The trailer setting is the colour; the institutional discipline of the pitch is the lesson.

📦 Public Launch and the Long Crawl (2015–2017)

Airtable launched publicly in March 2015, debuted on Hacker News, and closed a $3 million seed round on February 25, 2015 led by Caffeinated Capital with Freestyle Capital, Data Collective, and CrunchFund. A $7.6 million Series A from CRV (Charles River Ventures), with Ashton Kutcher participating, followed in May 2015.

The product evolved fast: API and embedded database in April 2015, Forms in July, Zapier integrations and the Barcode field in August, an iOS redesign in December. In 2016 came Calendar view (April), the iPad app (July), the Gallery view plus native macOS and Android apps in September alongside the 2 million bases milestone, then Kanban view in November. In September 2017 the team launched Airtable Universe, the community gallery of templates and apps that became one of the most enduring sources of organic discovery.

The First Round Review interview with Andrew Ofstad captures the organic adoption pattern that VCs underestimated:

"I visited WeWork. I looked around and everybody's computer monitor had Airtable open. And I was like, 'Oh my God, this is actually a thing.'"

— Andrew Ofstad, First Round Review

The land-and-expand pattern looks like this:

By 2017 the seed pattern was clear enough that Airtable bought billboards across the SF Bay Area. The thesis was now visible: the wedge wasn't displacing Excel — it was creating a category of internal apps that had never existed, one creator at a time.

The wedge succeeded because Liu refused the maximalist pitch. Excel-as-numerical-tool stayed Excel; Airtable only competed for the makeshift-database swath where you couldn't attach files to a cell, couldn't save filtered views per-team, couldn't link two tables together. Defining the territory you don't want is what made Airtable's territory easy for a non-developer to understand.

🧱 Becoming a Platform: Blocks → Apps, Automations, Sync (2018–2020)

On March 14, 2018 Airtable raised a $52 million Series B led by CRV and Caffeinated and simultaneously launched Airtable Blocks — branded by Business Wire as "a powerful DIY software creation platform for non-coders." Blocks were the first signal that Airtable wasn't a fancy spreadsheet — it was infrastructure.

Eight months later in November 2018 came the Series C: $100 million at a $1.1 billion valuation, led by Benchmark and Thrive Capital. Airtable was officially a unicorn, and the Hacker News thread logged the cultural reaction — half congratulations, half "this is a glorified spreadsheet."

The platform expansion accelerated in September 2020, when TechCrunch covered the Series D: $185 million at roughly $2.585 billion. The same announcement shipped:

• Apps (Blocks rebranded)
• Automations (triggers + actions, no-code workflow logic)
• Sync (data flows from external sources)
• Marketplace (third-party Apps distribution)

Suddenly Airtable wasn't a database with views — it was a database, with logic, with sync sources, with a marketplace of extensions. The three-layer mental model was now a product reality.

That sequence is the canonical "Airtable approval workflow" that the 9x crash course walks through in 25 minutes. Two automations plus one interface plus two tables equals a working internal app — built by one person in an afternoon.

🚀 The Decacorn Era and Interface Designer (2021)

Three rounds in fifteen months. March 15, 2021: TechCrunch reported the Series E at $270 million led by Greenoaks Capital with WndrCo (Jeffrey Katzenberg), Caffeinated, CRV, and Thrive — post-money $5.77 billion.

November 9, 2021: the Interface Designer beta opened to all users, via the Airtable blog. For the first time, the no-code app surface was a first-class shipped product, not a marketplace block.

December 13, 2021: the Series F at $735 million led by XN, with Salesforce Ventures, Franklin Templeton, JPMorgan Growth, MSD Capital (Michael Dell), Silver Lake, T. Rowe Price and existing investors. Post-money $11.7 billion. Bloomberg framed the round as Europe expansion fuel; CNBC framed it as low-code's coming-out party.

In Liu's Upfront Summit talk weeks later, he was explicit about the durable-growth philosophy:

"It's the outer years of durable growth that really matter that drive valuation. Growing 100% for one or two years but expected to flatline is much less valuable than growing 70-80% with belief you can sustain that 3-4-5 years out."

— Howie Liu, Upfront Summit 2022

In early 2021 Liu also brought in Peter Deng as Chief Product Officer (full-time December 2020 after an advisor period). Deng had ten years at Facebook on News Feed, Messenger, Groups, and Events, then was the first Head of Product at Instagram, then Head of Product at Oculus, then Head of Rider at Uber. His framing on the Airtable Table Talk podcast became the canonical articulation of the platform thesis:

"Every bit of functionality we build, we think of it as building a Lego block. If we execute well over 5–10 years, customers can build a lot of different kinds of software, not just collaborative applications."

— Peter Deng, Airtable Table Talk #15 (2022)

Three pillars for the year: Data, Alignment, Simplicity. New customers asked "how do I get data in?" Mature customers asked "how do I connect key pieces of multiple workflows?" The PRD template at Airtable opened with a single question: what customer problem are you trying to solve?

🧠 Inside the Three-Layer Mental Model

The three layers — Data, Automations, Interfaces — are the canonical Airtable mental model. Every external tutorial teaches them in this order, and every internal product launch fits into one of the three.

Layer 1: Data — typed, relational, view-aware

Field types are the most underrated piece of Airtable's architecture. The type determines what data is allowed: a date field rejects text, a single-select field is restricted to defined options, a link-to-another-record field threads bidirectional relationships and auto-creates the reverse field on the other side.

View types sit on top of the same underlying data. One table, many ways to see it:

Layer 2: Automations — triggers and actions

Automations have two halves. Triggers specify when a flow runs; Actions specify what happens.

Every record in Airtable has a RECORD_ID() — a unique identifier that automations pass around to update the right record. You can surface it as a formula field for debugging:

RECORD_ID()    →  "rec0aB1cD2eF3gH4i"

The conditional branch is the unlock. Peter Deng called it out as one of the "big" features he was unembarrassed about shipping: "I've never written a program that didn't have an if/else statement." Loops, error handling, and branching turned automations from one-shot triggers into real workflows.

Layer 3: Interfaces — the no-code app surface

The Interface Designer is what the end user actually touches. Layouts come pre-built (List, Dashboard, Kanban, Record Review, Form). The right panel configures whatever element is selected. Permissions are managed at the interface level — you build a public campaigns interface and a restricted admin interface, then scope them to specific user groups.

┌─────────────────────────────────────────────────────────┐
│ Pages    │  ┌────────────────┐ │  Right Panel           │
│ ▸ Home   │  │  CANVAS        │ │  ┌──────────────────┐  │
│ ▸ Admin  │  │                │ │  │ Element settings │  │
│ ▸ Inbox  │  │  [Card]        │ │  │ Source: Tasks    │  │
│ ▸ Form   │  │  [Card]        │ │  │ Filter: due ≤ 7d │  │
│          │  │  [Chart]       │ │  │ Group by: status │  │
│          │  └────────────────┘ │  └──────────────────┘  │
└─────────────────────────────────────────────────────────┘

The killer pattern: filter records to "the current user." A single interface scopes itself per viewer, so every team member sees only the records they're assigned to. By default everything is read-only — you toggle user actions on per element to allow editing, creating, or deleting records.

Side-by-side: the three layers

🤝 Customer Spotlight: Who Actually Builds on Airtable

Airtable's customer list reads like a tour through every category of internal app. In 2026 the homepage shows logos for OpenAI, Anthropic, Harvey, Amazon, Walmart, Netflix, Google, Airbnb, American Express, and HBO. The historical roster includes Tesla, Adobe, Atlantic Records, Condé Nast Entertainment, Shopify, Time, Autodesk, and Intuit.

The viral wedge that Andrew Ofstad described — "head of production goes from Netflix to next company, brings Airtable" — is what made the customer list look so industry-spanning. The pattern repeats: a single power user adopts the tool inside a function, the function adopts it across the org, the org adopts it across departments, and finally an enterprise contract gets signed.

A 2026 highlight: Airtable's Shopify automation piece runs live, syncing inventory, orders, and customers bidirectionally across the e-commerce stack.

🤖 Enter AI: From Airtable AI to the First Cracks (2023)

On May 10, 2023 Airtable announced its AI launch — AI Field Type, AI Component Library, "Apps by Airtable," Verified Data, Two-Way Sync, and a Services Partner Program. The beta program opened June 27, 2023.

The launch quote captured the moment:

"With AI breakthroughs that are capable of a broad range of reasoning and creative work, every form of knowledge work faces imminent transformation."

— Howie Liu, May 10 2023

The product worked. The strategy didn't, yet. Throughout 2023, Airtable's growth team was rebuilding onboarding under Head of Growth Lauryn Isford — who later joined Notion. Her contrarian framework, shared on Lenny's Podcast, still informs how Airtable thinks about activation:

"An activation rate that falls in a lower percentage range — maybe 5 to 15 percent for most companies — is better than one that falls in a higher percentage range, because there's likely much higher correlation with long-term retention."

Isford rebuilt onboarding around three buckets: an immersive guided wizard (highest impact), learning-style segmentation (technical-fluent vs visual learner, instead of role-based), and ongoing education via a design pattern internally nicknamed "the Mole." Result: roughly 20% lift in activation rate over 12 to 18 months.

But the macro environment was about to bite.

💀 The Two Layoff Rounds and the C-Suite Reset (2022–2023)

Airtable executed two layoff rounds nine months apart, cutting roughly 491 people in total — by some employee accounts on Glassdoor, close to 47% of staff over nine months.

December 8, 2022 — Round 1:

• ~254 employees (~20% of staff)
• Three execs depart: Seth Shaw (CRO, joined Nov 2020), Peter Deng (CPO, joined Dec 2020), Johanna Jackman (CPeopleO, joined May 2021) — all stayed on briefly as advisors
• Airtable retained the full $735M Series F
• Pivot framing: "narrowly focused mode of execution" targeting 1,000+ employee enterprises
• (TechCrunch coverage)

September 14, 2023 — Round 2:

• 237 employees (~27%)
• Affected: product and SMB-sales primarily
• Severance: 16 weeks pay, 2 months accelerated equity vesting, 6 months of health care
• Liu to Forbes (via SF Standard): "It's a sickening feeling. I made the decisions that got us here."

The 2024 enterprise pivot showed up in the product structure too. The plan ladder simplified to Free / Team / Business / Enterprise Scale — the "Plus" plan was eliminated. HyperDB was announced September 26, 2024, built to sync 100M+ records from Snowflake, Databricks, and Salesforce. ProductCentral launched October 16, 2024 as a vertical solution with JetBlue and BlackRock cited as anchor customers (BlackRock attributing "2x product velocity" to it).

In January 2024 Airtable also acquihired Airplane, an internal-tooling startup, sunsetting the original product by March 2024.

📰 The "Airtable Is Dead" Moment

Sometime in 2024, a viral tweet declared Airtable dead. Howie Liu's framing on Lenny Rachitsky's podcast, captured in a framing tweet by Lenny, recounts the dynamic precisely: a writer published numbers that were "wrong by a strong multiple" on revenue and growth rate, and the take spread before any correction could land.

"A lie gets around the world some number of times before truth has even has time to get out of bed."

— Howie Liu, Lenny's Podcast (Aug 2025)

That experience — the embarrassment plus the realization that the AI wave had moved past the company — became the catalyst. By the time of Liu's Lenny appearance in August 2025, the product strategy had already turned over.

Verification note: agents researching this article could not source the original CB Insights tweet or a specific All-In podcast segment. Lenny's framing tweet is the canonical reference; the article does not assert specifics that aren't sourced.

🛠️ Cobuilder: The Fastest-Adopted Feature in Company History (2024–2025)

Cobuilder was Airtable's first major AI bet and, internally, the fastest-adopted feature the company had ever shipped. Released in 2024 (Airtable newsroom), it let users describe an app in plain language and watch the database, the views, and the basic interface assemble themselves.

The output was a real Airtable base, not a synthetic mockup. That detail mattered: the agent wasn't generating code that someone had to deploy — it was generating Airtable primitives that the user could immediately edit, extend, and ship.

Cobuilder set the architectural pattern for everything that came next. Instead of having an LLM write a database from scratch (the failure mode of pure vibe-coding tools), have it compose Airtable's existing primitives. The agent gets reliability for free; the user gets editability for free.

That insight became the wedge for the AI-native relaunch the next year.

🔁 The Refound: Howie Liu's IC CEO Era (2025)

In June 2025, Liu reframed Airtable publicly as a refounding moment. The August 2025 Lenny Podcast episode — How we restructured Airtable's entire org for AI — laid out the philosophy. The frame had three parts:

1. The IC CEO — CEOs becoming individual contributors again, hands on the product
2. Refounding every product — AI is not a one-time form factor shift; every model release implies new UX
3. Fast-thinking vs slow-thinking org — two halves with different operating cadences

"Every software product, in my opinion, has to be refounded."

— Howie Liu, Lenny's Podcast (2025)

"It's now hard to taste the soup without participating in at least some part of creating the soup."

— Howie Liu, on the IC CEO frame

Liu's daily workflow shifted radically. Standing one-on-ones got cancelled by default. The barbell: longer in-person lunches every one to two months for genuine relationship building, plus topical timely meetings driven by real "alpha." A new standing meeting — a weekly AI execution sprint check-in — replaced most of the recurring calendar.

The Dan Shipper signal he cited: what predicts a company successfully adopting AI? Does the CEO use ChatGPT or Claude daily? Liu's answer: hourly.

His own usage included LLM map-reduce over a year of sales call transcripts to extract product, marketing, and positioning insights — costing hundreds of dollars per run, with equivalent consulting work running into millions. Liu has stated he is one of Airtable's largest individual inference-cost users globally.

The "go play with every AI product for a week" sprint

The sharpest piece of practical advice Liu gave in the Lenny conversation was the permission slip he started giving people on the EPD org:

"If you want to cancel all your meetings for like a day or for an entire week and just go play around with every AI product that you think could be relevant to Airtable, go do it."

— Howie Liu, Lenny's Podcast (Aug 2025)

The logic ladders out of the chef metaphor: you cannot understand new ingredients from a recipe card, you have to actually cook with them. AI is something you have to play with — both the packaged-up product layer and the underlying primitives reachable via API or chat — to know what new dishes are even possible.

That permission is what most product orgs are still failing to grant in 2026. The companies that compound do, the ones that stall don't.

That cycle — identify, build, ship, measure, repeat with the new ingredient — replaced the slower roadmap cycle. As of 2025, Liu reported that roughly half of Airtable's EPD organization is now on AI capabilities.

On the Greg Eisenberg podcast in April 2026, he described his coding workflow:

"I have like 30 different Claude Code instances running in parallel and each one is coupled up to a browser fully autonomous. It can go and get other agents to comment on any PRs it creates."

That's frontier mode. And it's the reason a 14-year-old company is shipping on the same cadence as AI-native startups.

🏗️ Fast-Thinking vs. Slow-Thinking: How Airtable Restructured for AI

The org evolved through three shapes. The current one — fast/slow — was inspired by Daniel Kahneman's Thinking, Fast and Slow.

The fast-thinking AI Platform group ships major capabilities on a near-weekly basis. Each release has to be jaw-dropping — not incremental. Who succeeds on the fast team? People with high autonomy and entrepreneurial instincts (literal former founders included), comfort thinking full-stack across technical layers and UX, and a relish for openness rather than feeling overwhelmed by it.

The slow-thinking group handles deliberate bets requiring premeditation. HyperDB — Airtable's data store now handling multi-hundred-million-record datasets — is the canonical example. You cannot ship HyperDB as a hacky one-week prototype. The slow group owns the infra that turns initial seeds of adoption into durable enterprise growth.

The two halves complement each other. Fast equals top-of-funnel excitement, AI-tourist traffic, new use cases. Slow equals retention, expansion, and the durable revenue base. Liu's diagnosis of pure-AI-native startups: wide top of funnel but weak retention. Airtable's slow half is the answer to that.

The full reorg arc — three shapes in five years

The fast-versus-slow split is the third reorg, not the first. Liu walked through the entire sequence on the Lenny podcast:

   Reorg #1 (2020-21)              Reorg #2 (~2022-24)             Reorg #3 (2025-)
   FEATURE TEAMS                   BUSINESS-UNIT PILLARS           FAST + SLOW THINKING
   ─────────────                   ─────────────────────           ────────────────────
   Search team                     Enterprise pillar               AI Platform (fast)
   Mobile team                     Teams / NUX pillar              Deliberate Bets (slow)
   Sync team                       AI pillar
   etc.                            Solutions pillar
                                   Infra pillar
   ↓                               ↓                               ↓
   Each remit = 1 feature.         Each pillar = 1 BU outcome.     Two cadences.
   Incremental by definition.      Crossfunctional bets possible.  One-cohesive product.
   Hard to plan org-wide bets.     Still slow vs cursor / surf.    Ships near-weekly.

Each shape solved a specific problem of the prior one. Feature teams couldn't make holistic bets — every team's remit was an incremental improvement to one surface. Business-unit pillars unblocked holistic bets within a customer segment but still left the road map fragmented across pillars when AI-native peers were shipping a single coherent product near-weekly. Fast-plus-slow is the answer Liu landed on for the agent era.

The quality bar for the fast group is not "ship something" — it's higher than that.

"Each of them should be like truly awesome value, right? Like you should drop your jaw at like how awesome it is to use this new capability in Airtable."

— Howie Liu, on the AI Platform team's bar (Lenny's Podcast, Aug 2025)

That bar is the reason a 14-year-old company is now releasing capabilities at the same cadence as a one-year-old startup. Speed without taste is noise; speed with taste is what compounds.

🧪 The Refound Test: "Useful Building Blocks or a Buyer"

The piece of the Lenny conversation that operators have been quoting back at each other since August 2025 is Liu's binary test for any pre-AI software company:

"If you were literally founding a new company from scratch with the same mission, how would you execute on that mission using a fully AI-native approach? If you can't, then you should find a buyer — and then if you really care about this mission, like go and start the next carnation of it."

— Howie Liu, Lenny's Podcast (Aug 2025)

The test has two layers. The first is whether the existing primitives are useful building blocks you can leverage — or legacy assets you would be better off without. The second is whether you have the honesty to introspect and act on the answer.

Liu's own answer to the test for Airtable: the no-code primitives — typed fields, link-to-record, automations, interfaces — do accelerate an AI-native build, because the agent doesn't have to generate every layer from scratch. That belief is the architectural premise behind Omni, Superagent, and Hyperagent. If Liu had concluded the primitives were dead weight, the Lenny quote is unambiguous about what he says he would have done: sell, then start over.

Most pre-AI SaaS companies still owe their teams that conversation. Few have had it.

🦾 The AI-Native Relaunch and Omni (June 24, 2025)

Airtable's relaunch announcement on June 24, 2025 unified the Cobuilder and Assistant features into a single conversational surface called Omni. Field Agents launched alongside as a new agentic primitive that runs inside individual fields. AI capabilities and credits became included in every plan — including Free.

The core Omni architecture is the dodge. Pure vibe-coding business apps from scratch hits known failure modes:

• Unreliable, bug-prone code
• Data and security issues
• Context collapse — the agent can't manage all the code as the app grows

Airtable's primitives — a CRUD interface on a data layer, beautiful by default — let the agent manipulate a working app without writing every layer from scratch.

"When the cost of making apps goes to zero, how many more apps will we see made?"

— Howie Liu, AI-Native Airtable launch letter (Jun 24, 2025)

Omni's pricing is structured to encourage adoption. App building and iteration — no extra cost. Data analysis Q&A — 10 credits per response. The bet: meet users where they are, charge for the heavy lifting.

The form-factor ladder: Copilot → Cursor → Composer → Omni-style workspace

Liu's clearest framing of why the Omni surface exists is a four-step ladder of how model capability has shaped UX form factor since 2022:

"We now made our agent the default way of doing everything in Airtable… the Airtable app as you know it is almost like an artifact that's manipulated by, and tool-used by, the agent."

— Howie Liu, Lenny's Podcast (Aug 2025)

That sentence is the architectural claim of the entire AI-native relaunch. The product surface didn't add a chatbot — it inverted the relationship. The conversational agent is now the canonical entry point, and the classic Airtable canvas is something the agent edits on the user's behalf. The same inversion is the design premise of Taskade Genesis: one prompt produces the database, the automations, the interface, and the agents in a single workspace.

The form-factor ladder also explains why "throw a chat sidebar on it" was never going to be enough. Each model generation implies a novel form factor to capitalize on its new capability — and the company that ships the right form factor first owns the era.

🧠 The DeepSky Acquisition: How Gradient Became Airtable's AI Engine (Oct 13, 2025)

Upstarts Media broke the news on October 13, 2025: Airtable had completed its largest acquisition in company history, buying DeepSky (formerly known as Gradient, which had raised approximately $40 million in venture funding). The deal brought 12 staff plus founders Chris Chang (CEO), Forrest Moret, and Mark Kim-Huang.

The bigger move: David Azose joined as Airtable's new CTO. Azose had previously led engineering for ChatGPT Business Products at OpenAI, after stints at DoorDash, Uber, and Microsoft. The Seattle-based hire signaled that Airtable's AI ambitions were no longer a roadmap item — they were a separate engineering org with its own leadership.

DeepSky operates semi-independently inside Airtable. Chang reports to Liu directly. The organizational signal: this acquisition wasn't bolted onto an existing team — it was given autonomy to ship, fast.

The first major output of that autonomy landed three months later.

🌐 Superagent: First Standalone Product in 13 Years (Jan 27, 2026)

On January 27, 2026, Airtable launched Superagent at superagent.com — its first standalone product brand in 13 years. TechCrunch covered the launch with a sharp framing: Airtable's secondary-market valuation had fallen $7.7 billion from peak, but Liu thought Superagent was just the warm-up.

Superagent is a multi-agent research and analysis system. A coordinator agent deploys parallel specialists across different data sources (FactSet, Crunchbase, SEC filings, plus your own Airtable bases). The output isn't a wall of text — it's interactive matrices, comparison cards, maps, and structured deliverables you can edit and re-prompt.

"You're not prompting an AI. You're orchestrating a team. That's the difference."

— Howie Liu, Superagent launch (Jan 27, 2026)

Superagent uses frontier models from OpenAI, Anthropic, and Google. Pricing tiers run from $20/month per user up to $200/month for power users, with bundled inference credits (per TechCrunch). VentureBeat's deep dive emphasized the execution-visibility angle: every step the multi-agent system takes is inspectable and auditable, not a black box.

For the Airtable category, this was the clearest "we are back" signal in three years.

⚡ Hyperagent: Cloud Agents with Their Own Compute (Early 2026)

If Superagent was Airtable saying "we ship standalone products again," Hyperagent was Liu saying "the agent surface deserves a separate brand on a separate domain." Hyperagent.com launched as a fully separate site, with its own visual identity and pricing.

Hyperagent is a turnkey agent platform where every session gets its own isolated, full computing environment in the cloud. Real browser. Real shell. Real filesystem. Image and video generation. Mapping. Data warehouse access. Hundreds of enterprise integrations. Skill-learning compounds across sessions. One-click Slack deployment makes any agent deployable as a Slack-native coworker.

Homepage copy (verbatim)

• Hero: "Weeks of work. Overnight."
• Core: "The system of agents that does real work, learns how your organization operates, and deploys across your entire team."
• Anti-positioning: "Not a chatbot. Powerful cloud agents with a full computing environment."

Featured demos with token costs

Those numbers are the citation capsule. The reframe Liu pushed on Greg Eisenberg's podcast:

"Think about the opportunity cost of my time. That's the real reframe — human equivalent time cost vs. '$150 sounds expensive vs. a $10/month sub.'"

A $150 inference run that produces a board memo — Liu's actual recent example — replaces consulting work that would cost millions and take weeks.

Launch incentives

• Founding 500: $10 million in inference grants
• First 1,000 signups: $1,000 bonus inference credits + 2.5x cost subsidy on top-tier Claude and other frontier models, up to $15,000/month, for one year
• Greg Eisenberg podcast community: separate $1,000 to first 1,000 builders ($1M total community pledge)

How Liu positions Hyperagent vs. the field

The fleet command center

Once you build many agents, Hyperagent surfaces them as a fleet — content marketer, market researcher, customer email responder. Each plays a role and can be deployed to Slack one-click. Each accumulates memory and skills over time.

Rubrics: the eval-as-first-class-primitive differentiator

This is the piece Liu was clearest about as the unlock that separates Hyperagent from Manus or Perplexity Computer:

"If you're using Manus, who is the judge? The judge is you, the human. It's not a top-tier frontier model, it's a huge gating factor."

Hyperagent rubrics let you define what good looks like for a task, pin a rubric to an agent, have an LLM-as-judge score every run on the dimensions you care about, see trend lines over time, and auto-suggest model downgrades to a smaller model (~5x cost reduction) when score doesn't drop.

Skills as the primitive

"Skills are the most important concept and primitive in the frontier agents world. The models are generally intelligent enough — like Einstein. He may not know real estate, but if you give him a playbook, he figures it out."

Skills are composable, interactively created, and evergreen — they improve over time via automatic learnings or interactive correction. Every agent run accumulates new memories, skill suggestions, system-prompt suggestions, and tool-access suggestions. The fleet is self-improving by construction.

This is what Liu means by "the agent surface deserves a separate brand." Hyperagent isn't an Airtable feature — it's a different product category, with its own IA, its own pricing, its own URL.

The Hyperagent skill lifecycle

Skills compound across runs. Memory accumulates. The fleet self-improves.

Liu's framing: agents converge on human-shaped roles for the same reason humanoid robots converge on humanoid form — infrastructure is built for human ergonomics. With agents the constraint is the physics of context windows. We won't get infinite context, so we partition agents the way we partition humans: into roles in a company.

🛍️ The 2026 Product Stack — Full Lineup

The shape of the company is now visibly two-product: the legacy platform (Airtable + HyperDB + Omni + Field Agents) and the standalone agent brands (Superagent + Hyperagent). DeepSky-the-org powers both.

💰 The Funding History

Total raised: approximately $1.36 billion per Airtable's own Series F newsroom, restated as ~$1.4 billion in TechCrunch's January 2026 piece. Cash retained: ~$700 million. No Series G announced as of mid-2026.

The arc is classic decacorn. The Series F was opportunistic — Liu told Upfront Summit 2022 that they hadn't touched the Series E yet and raised because the round came to them, not because they needed cash. That decision now looks prescient: $700M of dry powder funded the entire AI refound without a dilutive Series G.

👥 The People Behind Airtable

The exec roster turned over significantly in late 2022 and again in late 2025. The current leadership shape:

The throughline: founders still run the company. Liu remains CEO, Ofstad remains CPO, Nicholas remains an active co-founder. The 2025 hires reinforced rather than displaced.

🥊 Airtable vs. The Field (2026)

Airtable invented the low-code business-app category but the landscape around it has multiplied. Each adjacent player took the same primitives and weighted them differently.

The simplification: every player took some subset of the three layers and added one bolt-on. Airtable started with the strongest data layer and earned the right to add agents. Taskade Genesis takes the same three-layer pattern and adds agents as a first-class fourth layer — apps that come pre-loaded with their own memory, agents, and automations.

For deeper category dives, see the complete history of Notion AI, the history of Monday, and the history of ClickUp.

🎨 The Under-Merchandised AI Problem (Liu's PM Critique)

The most counterintuitive product critique Liu made in the Lenny conversation was self-directed at the entire AI-product category, ChatGPT included. His point: most AI products today are under-merchandised. The capabilities are real, but the visual metaphors and affordances on top of them are weak — a giant blank chat box is the dominant pattern, and it tells the user almost nothing about what the model can actually do.

"Most of [AI's awesome capabilities] are really under-merchandised, and there's like very poor visual or otherwise metaphors or affordances given to users to help represent or understand what those underlying capabilities are."

— Howie Liu, Lenny's Podcast (Aug 2025)

The implication for product teams in 2026 is concrete. The roles inside an AI-product team need to flatten:

"There's a strong advantage to any of those three roles who can kind of cross over into the other two. As a PM, you need to start looking more like a hybrid PM-prototyper who has some good design sensibilities."

— Howie Liu, Lenny's Podcast (Aug 2025)

Same insight, different domain: Nick Turley, Head of ChatGPT at OpenAI, has framed the related discipline on the engineering side as asking "is this maximally accelerated?" of every initiative — and Turley adds the second half: with AI, you often don't know what people want to do with it until it's out. The two beliefs combine into a single operating principle for the agent era — ship fast, ship visual, let usage tell you what to merchandise next. Airtable's fast-thinking org ships near-weekly precisely because that loop is the unlock.

For Taskade Genesis, this is the same reason every layer ships with first-class affordances — the database, the agents, the automations, and the published app are all visible in one canvas the moment you describe them. There's no blank chat box pretending the capability doesn't exist.

🧬 Why Workspace DNA Beats Database + Bolt-Ons

Airtable's three-layer pattern — Data, Automations, Interfaces — is the canonical low-code shape. Taskade Genesis takes that shape and adds Intelligence as a built-in layer. The mapping is clean:

The key difference: in Airtable the AI layer is bolted onto the existing primitives. In Taskade Genesis, Memory feeds Intelligence, Intelligence triggers Execution, and Execution writes back to Memory — a self-reinforcing loop where every layer was designed alongside agents from day one.

Airtable's argument is "we're already the trusted database; AI should ground on us." Taskade's stronger version: the database, the agents, and the automations are one workspace. Memory feeds Intelligence which triggers Execution which writes back to Memory — no MCP round-trip required.

The lesson Rob Bridner articulated on the BuiltOnAir podcast — that AI agents need a single source of truth and visual feedback when they act — is exactly the gap Taskade Genesis closes by construction.

"Visual feedback and confirmation" — Rob's pain point with autonomous research agents is the gap Taskade Genesis closes natively. Apps built with Taskade Genesis surface what the agent did, what records exist, and let humans inspect and edit at every step.

🚀 Full Taskade Genesis Capability Surface (the apples-to-apples table)

For builders deciding between Airtable and Taskade Genesis in 2026, here is the full surface — every Taskade Genesis primitive that maps to (or extends) the Airtable model:

The architectural difference Liu names — "every session gets its own isolated, full computing environment in the cloud" — is what Hyperagent uses to dodge context collapse on its own domain. Taskade Genesis dodges the same failure mode in a different way: by composing the Workspace DNA primitives (Memory + Intelligence + Execution) inside one workspace, so the agent never has to re-collect context that already lives next to it.

Both paths work. Hyperagent is the right answer when you want a fleet of agents that go deep on enterprise integrations from their own surface. Taskade Genesis is the right answer when you want one workspace where the database, the agents, and the automations grow together — and the app deploys the moment you describe it.

📚 Further Reading

History posts in this series

• Anthropic & Claude AI Complete History — $380B, Constitutional AI, Claude Code Agent Teams
• OpenAI & ChatGPT Complete History — GPT-5, Stargate, the $500B race
• Notion AI Complete History — all-in-one workspace, $11B journey
• Monday.com Complete History — Work OS evolution
• ClickUp Complete History — one-app productivity
• n8n Complete History — workflow automation roots
• Lovable Complete History — vibe coding origins
• Replit Complete History — Ghostwriter to Agent
• History of Mermaid.js — diagrams as code

AI architecture & agent design

• What Is Agentic Engineering? — how AI agents reshape software
• Single-agent vs Multi-agent AI Teams — fleet patterns
• Agentic Workflows: Path Towards AGI — design patterns
• How Do Large Language Models Work? — transformers explained
• Vibe Coding vs No-Code vs Low-Code — category breakdown
• Code vs Runtime: Taskade Genesis Manifesto — they generate code, we generate runtime
• Agentic Workspaces — the Workspace DNA loop

Try Taskade Genesis

• Build a free Taskade Genesis app — one prompt, one live app
• AI Agents — custom agents with memory, slash commands, 22+ built-in tools
• Automations — 100+ bidirectional integrations, durable execution
• Community Gallery — clone real Genesis apps in one click
• Free Airtable alternative — Workspace DNA, included
• Learn: Projects & Databases — the Memory pillar deep dive

🛣️ Try It Yourself

Airtable invented a category and is now refounding itself for the AI era. The architecture lesson — don't have the agent write everything from scratch; have it compose existing primitives — is the right answer to the context-collapse problem.

Taskade Genesis takes that lesson one step further. Memory, Intelligence, and Execution are one workspace. Agents come built-in. Automations come bidirectional. One prompt produces the database, the agents, and the deployed app — all wired together, all live, all editable.

One prompt. One app. Diagrams that live. Memory that feeds intelligence that triggers execution.

Try Taskade Genesis free →

Frequently Asked Questions