Howie Liu's Founding 500 program is not a normal startup fund. It is a $10 million bet that the next generation of companies will be built around agents before venture capital fully catches up.
Howie Liu is taking the accelerator idea and bending it around AI infrastructure. The Airtable co-founder and CEO is backing 500 founders through Hyperagent's Founding 500 program, offering $20,000 in inference credits to each qualifying applicant, with $10 million committed across the cohort.
That matters because this is not a simple grant announcement or another founder-friendly branding exercise. It is a distribution move, an infrastructure subsidy and a market signal wrapped into one. If agents become the operating layer for small companies, the first founders who learn how to build with them will have an advantage that looks less like software adoption and more like company formation itself.
As Digg captured from Liu's May 22 X post, the program is aimed at founders building “agent-first businesses,” with Hyperagent positioned as the platform those builders will use to run autonomous, proactive agents. A separate Hyperagent community post says the offer gives the first 500 qualifying applicants $20,000 in inference credits, with a May 31, 2026 application deadline.
The details are important. This appears to be credit-based backing, not a $10 million cash fund. A Reddit post from a Hyperagent builder evangelist also says $200 unlocks the $20,000 credit allocation, so founders should treat this as subsidized compute and platform access rather than traditional seed capital. That distinction does not make it small. For agent startups, inference is one of the real costs of experimentation.
Liu is not coming at agents as a tourist. Airtable spent years making database-backed software easier for non-technical teams, and the company has recently pushed harder into AI-native work. In January, Airtable introduced Superagent, a standalone multi-agent product that Liu described as a system for coordinating specialist agents rather than prompting one assistant to work through tasks sequentially.
That background gives Liu a useful lens. Traditional investors can evaluate markets, teams and revenue. A founder who has lived through no-code, enterprise adoption and workflow automation may be better positioned to spot whether a young agent company is building a real operating model or just a thin wrapper around a frontier model. That is the difference that will matter as the agent market fills with confident demos.
Most founders in this category are not only asking whether an agent can complete a task. They are asking whether it can remember context, use tools, recover from failure, work across systems and produce an output that a customer will actually pay for. Hyperagent's pitch leans into that practical layer: agents with browser access, shell access, code execution, integrations and deployment into tools like Slack.
This is where the no-code lineage becomes useful. Airtable did not win by telling every business user to become a software engineer. It won by packaging structure, data and workflow into something teams could adopt without waiting for central IT. If agents follow a similar path, the winners may be the companies that make autonomous work feel manageable, inspectable and repeatable.
The accelerator model is changing
The Founding 500 also shows how seed-stage competition is shifting. Y Combinator still has the strongest startup brand in the world, but AI infrastructure companies now have another playbook. They can offer credits, community, templates, distribution and technical feedback before a founder ever raises a priced round.
That is attractive because the bottleneck for many agent-first startups is not always office space or pitch coaching. It is the cost of running enough experiments to know what works. An agent that browses, writes code, calls tools and loops through multiple attempts can burn through inference quickly. A $20,000 credit pool can buy a serious amount of learning for a small team.
There is also a strategic reason for Hyperagent to move early. If 500 serious builders start constructing companies on its platform, Hyperagent gets feedback from real edge cases before larger competitors can flatten the category. Some of those companies may fail, but the usage patterns will still teach the platform what agent-native businesses actually need.
For founders, the opportunity comes with a tradeoff. Building deeply around one agent platform can speed up development, but it can also create dependency. The smart teams will use the credits to validate workflows, customer demand and operating assumptions, while keeping a clear view of what belongs to the company versus what belongs to the infrastructure provider.
The bigger question is whether agent-first companies become a durable startup category or just a temporary label for companies using better automation. The answer will show up in revenue, not rhetoric. If small teams can serve enterprise customers with fewer employees, faster execution and lower coordination costs, then the company-building model really does change.
Liu's program is worth watching because it places that question in the hands of 500 builders at once. The next signal will not be the size of the announcement. It will be whether any of these founders turn subsidized inference into products customers cannot stop using.
Also read: MoonPay brings crypto buying inside ChatGPT. • AI startups are making ARR harder to trust. • Tencent's L2P makes pixel-space image generation practical again
