Airtable’s new Cobuilder product puts AI-generated apps in every worker’s hands — and has the $11.7 billion-valued startup back eyeing an IPO following painful job cuts.
Airtable CEO Howie Liu calls his new AI tool Cobuilder the company’s biggest release in five years.
Leigh Nile
For more than a decade, Airtable CEO Howie Liu has worked toward a mission to make building software easier — its tools able to turn data tables and document troves into apps big and small to save workers time. But while Airtable’s low-code tools have proven popular — 500,000 organizations have generated 50 million such apps so far — they largely required hours of time and technical proficiency.
Now, Liu believes Airtable has found the solution to its biggest bottleneck: generative AI. And with a new tool launched on Wednesday, Airtable Cobuilder, his startup is leveraging a combination of leading AI models and its own new tools to make spinning up an app as easy as typing in a simple text prompt.
“What might have taken many hours you can do in literally 30 seconds,” Liu told Forbes. “We see this as a big unlock to enable many more applications to be built.”
Cobuilder isn’t Airtable’s first foray into AI — the company began integrating AI tools within its app-building process last year and added more capabilities in March — but it’s the first to lean on AI to do all the heavy lifting, not just speed up one part. Airtable generates the app by analyzing the user’s role, company and query and making a series of API calls to a large-language model — right now, mainly OpenAI’s GPT-4 model family — with its own improved, more specific prompts.
A visual effects manager at Netflix, for example, could tell Cobuilder their job title, and it might suggest an app to keep the scheduling and relevant files for an upcoming release all in one spot, pulling in info from calendars, Dropbox and Google Sheets. Users can also ask for more specific tools: An ecommerce seller could request an app that can track all shipments of a certain product, factoring in shipping costs and time. The result: what often looks effectively like a dashboard or interactive table, with the relevant variables (orders, due dates, customs waits) presented in a straightforward interface to tweak or toggle on and off.
Eventually, the company will provide the option for companies to include their own data within the AI generative process, but for now, users must manually hook it up after Airtable has returned back its app attempt. (Non-customers can also try launching apps on the company’s website.) Airtable’s software doesn’t train off those generated apps themselves — it does use meta data, such as whether the customer asked it to try again — nor does it share customer data with OpenAI, Liu said.
While the output doesn’t take much more effort for a user, the process to return a functioning app is more complex under the hood than, say, asking a question of OpenAI’s ChatGPT. “We have to do a lot of stuff behind the scenes, because the models are not omnipotent,” Liu said. And unlike with fact-based requests, Cobuilder’s responses aren’t so simply right or wrong. “There might be 50 different viable ways to implement an app and how it’s structured, and some will do a better job than others.” Airtable’s AI tools absorb user feedback — not the generated outputs themselves — and self-corrects over time for better ways to map an app to a type of request.
Testers have built thousands of apps using Cobuilder in the past few weeks, but the company declined to name specific users. It has previously named Amazon Web Services as an early Airtable AI adopter, and claimed to work with a large player in both law and media. Aetna, Nike and Walmart are other notable Airtable customers. For regulated industries like banking and healthcare, Airtable is also developing tools for managers like chief information officers to easily label which data and workflows should be available for employee app use, “blessing sources of truth” for say, any app looking to use a company’s product roadmap, or upcoming marketing materials.
Airtable’s AI push represents a new phase for a company that careened in recent years from buzzy startup darling to cautionary tale. Founded by Liu and two others in 2012 and launched the following year, Airtable echoed popular design software startup Figma in taking years to ship a product, then finding success because people flocked to the product organically — a beacon of, in startup lingo, “product-led growth.”
By 2018, venture capital firms including CRV, Benchmark, Thrive Capital and Coatue had backed Airtable; as the then 30-year-old leader of a $1.1 billion-valued unicorn, Liu became the subject of glossy profiles. By December 2021, investors valued Airtable at $11.7 billion, giving it one of tech’s loftiest price tags. A few months later, the company slotted in at No. 6 on the Forbes Cloud 100 list in 2022.
What an AI-generated app looks like in Airtable.
Airtable
But as the free-flowing capital of a venture bull market fueled in part by low interest rates abruptly ceased flowing that year, Liu was caught flat-footed. Employees paid the price for bloated hiring and unrealistic visions of growth as Airtable laid off 20% of staff that December, then another 27% in September 2023, a combined nearly 500 people. “It’s a sickening feeling,” Liu said then. “I made the decisions that got us here.”
Reflecting a year later, Liu admitted Airtable had gotten “disoriented” by its own momentum. “The froth obscured the true nature and quality of pretty much every business that was in hyper-growth, hyper-funded mode,” Liu reflected. “Nobody had to put their cards down and show how good their business really was.”
During that period, Liu had dinner with Coatue investor David Schneider, the former chief revenue officer and president at ServiceNow, raised in the classic school of enterprise sales. Schneider asked Liu to name his top 25 customers and how they used Airtable; Liu couldn’t. The message was clear: hit the pavement. “He pushed me to meet with them, and I did,” Liu said on a recent trip to New York to do just that. “This is where different business models require a different style of CEO.”
Those customer meetings gave Liu and his team cause for hope even as morale suffered with the departure of colleagues and multiple executives. “Howie was like the architect of a house that never saw the finished house,” said Schneider. “So, he went and saw the beautiful homes being built on top of his architectural drawings.”
Airtable wasn’t close to running out of money — it still has about $1 billion in the bank, including all of its $735 million Series F raised in 2021, and most of its $270 million Series E from a few months before, raised to give it “financial freedom,” Liu said at the time — but the tightening stabilized its footing, Liu said, with positive cash flows. That done, it could reinvest in its product again — notably, Airtable AI. “It’s not just about cutting your way and optimizing,” said Vince Hankes, a partner at Thrive Capital who has also worked closely with Liu. “Ultimately, we’re in technology, a growth business, and you need to find a path to get back to that.”
For Liu, Airtable’s opportunity to expand its market through AI — both in terms of how it can be used, and by whom — has provided that answer. “We will be a winner in this new world order. If I didn’t believe that, I would probably have explored options for a sale,” he said.
Instead, Liu made acquisitions of his own that proved crucial in the development of Cobuilder: Airplane, which offered custom workflows for developers, and Balsa, which made a tracking tool for software projects. Airtable bought both mainly for their people, who became core members of the Cobuilder push. “We have a very clear point of view of what layer of the AI stack we want to play in,” Liu said.
That means Airtable won’t build its own models or devote resources to adapting existing ones. Cobuilder leverages OpenAI today, Liu said, but the company is “agnostic” to any one model provider moving forward. And while Airtable is keeping tabs on open-source models, such as Meta’s recently unveiled Llama 3.1 model, Liu said he remains skeptical about the need to move off OpenAI anytime soon, given its pressures to ship more models at competitive prices: “These things will get cheaper, better and faster.”
Investors, meanwhile, are glad he hasn’t called private equity firms just yet. More than half the 500 largest companies in the U.S. by revenue are paying customers, Airtable said; more than 25 customers spend at least $1 million annually, and several at least $5 million. The company has “hundreds of millions” of revenue and clears the ‘Rule of 40,’ a much-used benchmark for software businesses that looks for revenue growth and profit margin to combine to at least 40%, according to Liu, who added that Airtable’s numbers would place it in the top decile of public cloud peers on the BVP Nasdaq Emerging Cloud Index.
A public offering is back in discussion, although Airtable’s CEO said that he, like many of his peers, is “waiting to see how the markets evolve.” Airtable announces its quarterly results internally already for the practice, he added, so it can be “IPO capable at any point.”
Are you – or is someone you know -creating the next Afterpay or Canva? Nominations are open for Forbes Australia’s first 30 under 30 list. Entries close midnight, July 15, 2024.
Look back on the week that was with hand-picked articles from Australia and around the world. Sign up to the Forbes Australia newsletter hereor become a member here.
