Airtable Lost $7 Billion in Paper Value. Now It's Betting Everything on Agents.

Howie Liu stood in front of his employees last fall and told them something unusual: Their stock options, once priced at an $11.7 billion valuation, had collapsed to roughly $4 billion. And that was actually good news.

The Airtable founder reframed the destruction of two-thirds of his company's paper worth as a recruiting advantage. Equity, he explained, was now "much more attractively priced." The upside, should his bets pay off, would be considerable.

This week, Liu revealed what he's betting on. Superagent, Airtable's first standalone product in thirteen years, represents either the company's smartest pivot or its most expensive distraction. The timing is either brilliant or catastrophic. Every serious software company is racing to prove it can deliver on agents right now, and Liu just entered a sprint where OpenAI, Anthropic, and Google have a significant head start.

The pitch: stop prompting, start orchestrating

Liu draws a technical distinction between Superagent and its competitors. Most products marketed as "AI agents," he argues, are actually predetermined workflows with some language model calls sprinkled in. They follow scripts. You ask a question and get a script, not a strategy.

Superagent, according to Liu, uses genuine multi-agent coordination. Ask it about expanding your athleisure brand into Europe, and the system doesn't start searching. It builds a research plan first, identifying what needs investigation and surfacing dimensions you hadn't considered. Then it deploys specialized agents in parallel. One investigates financials. Another analyzes competitive positioning. A third reviews management and recent news.

Watch the demo and you see something different from a chatbot. Progress bars multiply across the screen as agents spin up simultaneously, each labeled with its task. The output assembles itself in real time: an interactive market analysis where you can click into demographic breakdowns, hover over competitive presence mapped as colored dots on a European grid, and filter expansion timelines by risk tolerance. "What if every person could have New York Times-quality data visualization built for every task they have?" Liu asked in a recent interview. The question sounds like marketing copy, but the deliverables are genuinely different from a ChatGPT summary you'd need to reformat for three hours.

The distinction matters because most "agents" are just chatbots with extra steps. They process your request, call an API, process again, call another API. Sequential. Predictable. Superagent's orchestrator maintains visibility over the entire execution sequence, including the initial plan, every step, and all sub-agent results. When a research task fails, the system knows what it tried and won't repeat the same mistake. That's the theory.

Building a rooftop pool while the plumbing backs up

Here's the uncomfortable part. Airtable's core database product has problems. Serious ones. Record caps frustrate power users. API calls fail unpredictably. Enterprise customers wait weeks for critical support tickets to get addressed. One Reddit user, paying $75,000 annually, posted this week that they can't even open some bases at 80% capacity. "Stop chasing share value and deliver the bloody product I pay for," they wrote. "Your product sucks. You have crap admin tools. Crap support. AI ain't gonna fix it."

Airtable knows this criticism exists. The company feels cornered: ignore AI and watch competitors eat your market, or chase AI and confirm every frustrated customer's suspicion that you've stopped caring about the fundamentals. The Superagent launch looks, from the outside, like a company adding a rooftop pool while tenants complain about backed-up pipes. Liu would argue he's doing both. The core business still employs more than 700 people and serves 500,000 organizations, including 80% of the Fortune 100. That business throws off cash. But cash doesn't fix the product-market tension. Angry customers don't care how much runway you have.

The technology itself comes from DeepSky, an AI agent startup Airtable acquired in October for an undisclosed sum. DeepSky, formerly known as Gradient, had raised $40 million and built a reputation for long-context architectures. The founding team now runs Superagent semi-independently. Airtable also brought in David Azose as CTO last fall, straight from leading ChatGPT's business products at OpenAI. These hires signal ambition. They also signal that Airtable's existing engineering leadership wasn't enough for what Liu wanted to build next. The company feels defensive about where it stands in the AI hierarchy, and defensive companies make expensive acquisitions.

What you're actually buying

If you're evaluating Superagent, ignore the multi-agent architecture for a moment. Focus on the outputs.

Superagent pulls from FactSet, Crunchbase, SEC filings, earnings transcripts. The premium data access matters more than the architecture. I asked it to evaluate Google as a three-year investment opportunity. Back came citations to specific earnings calls, a defensibility analysis against OpenAI and Anthropic, risk factors I hadn't thought to ask about. Wanted a brief on Wells Fargo's AI strategy before a pitch meeting? Their regulatory posture, recent investments with deal details, the pain points your product actually addresses. All formatted.

You don't turn these into presentations. They already are presentations. They are the presentations: interactive, filterable, visual. You could walk into a board meeting with one of these outputs and not embarrass yourself. That's the value proposition. The question is whether the premium data access and polished deliverables justify the subscription, which starts at $20 monthly and scales to $200 for power users.

Liu acknowledges only two competitors with comparable architecture: Anthropic's Claude and Manus, a newer entrant that Meta just acquired. Everyone else, in his telling, is building "LLM-powered workflows" disguised as agents. The word has become so diluted it means almost nothing. But architectural purity matters less than results. If a simpler, cheaper product delivers adequate output faster, nobody cares how many agents coordinate under the hood.

The wartime CEO

Liu calls his current approach "wartime" leadership. He once dismissed the term as needlessly violent but now considers it apt. "Being very fast on the draw to be able to adapt," he said recently, is "the most value-creative way to run things right now."

The valuation collapse didn't undermine the business itself. Airtable raised $1.4 billion total and still has half of that in the bank. The company continues generating cash. Liu has capital for strategic acquisitions and doesn't need to raise another round. The $7 billion evaporation affected investor returns and employee stock options. It didn't affect his runway.

Seven hundred million in cash buys you room to swing. Superagent could become the trillion-dollar market Liu keeps mentioning in interviews, or it could land in the same graveyard as every other agent product that shipped six months too early. Asked whether Superagent might eventually eclipse Airtable itself, Liu didn't dismiss the possibility. "Airtable will probably be larger for at least the near term than any new products that we do, including Superagent," he said. "But I also like being able to bet on Superagent. Optionality is a good thing."

The bet I'd make

Liu has the resources to run this experiment without betting the company. That changes the calculus. A desperate pivot from a cash-strapped startup would be a warning sign. A well-funded side bet from a profitable company with $700 million in reserves is different. He can afford to be wrong.

But the frustrated enterprise customers reveal something Liu's wartime framing obscures. The engineers building Superagent aren't fixing API reliability. The dollars spent on DeepSky aren't raising record limits. That's the trade-off nobody at Airtable wants to say out loud. The rooftop pool might attract new tenants, but the existing ones are watching their bathrooms flood.

My read: the architecture is interesting, the deliverables are useful for strategy and finance teams, and the timing is risky. Liu is probably right that most "agents" are just chatbots with extra steps. He's probably wrong that this distinction will matter to buyers in eighteen months, when everyone has shipped something comparable. The window for architectural differentiation is narrow.

Superagent will find customers among executives who need polished research outputs and don't want to format them. It won't save Airtable's relationship with the power users watching their bases choke at 80% capacity. Those are two different problems, and Liu is solving the more exciting one.

Claude Code Sub-Agents: Build Your AI Team

Claude Code’s new sub-agents let you build specialized AI assistants instead of one generalist trying to do everything. Each operates in its own context window, preventing pollution and delivering higher quality results for specific tasks.

Implicator.aiMaria Garcia

Multi-Agent AI Coding: The Hidden Token Economics

Anthropic’s multi-agent Claude Code lets developers run five AI assistants simultaneously. Early adopters report 10x productivity gains. The token bills tell a different story about who really benefits from autonomous overnight coding.

Implicator.aiMarcus Schuler

GitHub Agent HQ: Orchestration Over AI Agent Competition

GitHub isn’t building the best AI coding agent—it’s building the layer beneath all of them. Agent HQ bets enterprises value coordination over specialization. Platform strategies that extract value from ecosystem competition reshape markets more durably.

Implicator.aiMaria Garcia