AI Agents Are About to Get Bank Accounts

OpenAI launched GPT-6 Astra last week. There will be plenty of debate about benchmarks, capabilities and whether this gets us closer to AGI. The more useful question for me is economic: what can we build with it now?

Astra is another meaningful step toward AI that can perform increasingly complex, multi-step work across software, research, coding and computer use. OpenAI describes it as capable of carrying complex tasks from an initial request through to a finished result. For anyone building companies, this matters. The cost, speed and organizational structure required to launch a new venture are changing quickly.

I believe 5–10 exceptional humans, amplified by dozens of specialized AI agents, can build and scale a real business today that would have required a much larger organization only a few years ago. That isn’t a prediction about 2030. It’s an operating model we can start designing around now.

There is another shift happening at exactly the same time that makes this considerably more interesting. AI agents are starting to get access to money.

Software has historically been relatively static. We told it what to do, gave it rules and clicked the buttons. Agents are turning software into something much more dynamic: software that can reason, make decisions, take actions and increasingly interact with other software on our behalf.

The next step is obvious, and a little uncomfortable: give it money.

Until recently, most AI agents eventually hit a fairly obvious wall. They could research, recommend, generate and increasingly execute, but when they needed to buy something, procure a service or commit financial resources, a human generally had to enter the loop.

That wall is starting to disappear.

Payment networks and financial institutions are beginning to build infrastructure that allows agents to transact under delegated authority, with identity, permissions, spending limits and controls. We are moving toward agents that can search, decide, purchase and potentially transact with other agents.

The internet connected software. AI is giving software agency. Programmable money gives it economic power.

An agent that can reason and act is useful. An agent that can reason, act and allocate capital becomes an economic actor.

I’ve spent much of my career building companies and the last decade at Highline Beta working with large organizations to build new ventures. The traditional equation has been fairly predictable: identify a problem, assemble a team, build a product, get customers, allocate capital, find distribution and scale the organization.

AI is compressing almost every part of that equation.

I’m starting to see this directly through the vertical venture studios I’m involved with. Small teams can increasingly surround themselves with specialized agents across research, product development, coding, customer discovery, marketing, sales operations and administration.

The model isn’t really 5 humans + 50 agents. It’s a small core team with an increasingly elastic layer of machine intelligence around it.

I don't think there will be a standard human-to-agent ratio. That's a little like asking how many spreadsheets an analyst should have. Agents aren't headcount. They're leverage.

The org chart is still going to get weird. The agents won't need performance reviews, engagement surveys or an HR department.

More importantly, this changes how quickly you can test an idea, how much capital you need before finding product-market fit and ultimately what kind of company you need to build. As models improve, the amount of organizational capacity available to each human keeps increasing without headcount increasing at the same rate.

If something that historically required 30 people and $5 million can increasingly be built by 5–10 exceptional humans with an elastic layer of machine intelligence, why are we still building 30-person companies and raising $5 million?

Some of the answer is habit. Venture capital has also spent decades optimizing around a world where scaling a technology company meant scaling headcount.

AI breaks that assumption. The bigger shift isn't simply fewer employees. The unit economics of organizational capacity are changing.

I think we're going to discover that a lot of early-stage companies are both overstaffed and overcapitalized.

There is another implication I think is being underappreciated.

AI may reverse part of the traditional startup advantage.

Startups historically beat corporations on speed, focus and willingness to take risk. AI makes speed and technical execution dramatically cheaper.

None of this means putting 50 agents inside the existing corporate operating model suddenly makes it innovative. AI inside bureaucracy is still bureaucracy, just with better prompts.

Giving everyone an AI licence isn't an AI strategy either. Putting agents into the same approval processes, organizational structures and incentives doesn't fundamentally change the operating model.

The bigger opportunity is organizational redesign.

The mothership may actually be the wrong place to capture the biggest AI opportunities. Large companies have extraordinary assets. They also have organizational antibodies designed to protect the core business. Both things can be true.

That’s why I think the opportunity sits with independent corporate ventures that operate with startup speed and incentives while selectively accessing assets startups spend years and enormous amounts of capital trying to acquire: customers, proprietary data, distribution, capital, brand, domain expertise and regulatory permissions.

Now combine those corporate advantages with a small exceptional team and an elastic layer of machine intelligence. The venture building equation starts to look very different.

You get startup speed without starting from zero.

This is increasingly how I'm thinking about the vertical venture studios we're building and operating at Highline Beta.

This is where my corporate venture building thesis starts colliding with my Future of Finance thesis.

My friends at RBC aren't going to bank one of my AI agents tomorrow. They already have to explain businesses like mine to traditional credit risk teams. I'm not going to make their lives harder by adding autonomous software to the mix. Frankly, I'm not sure what the KYC form would even look like.

The agent doesn't care. It doesn't necessarily need a traditional bank account.

It needs financial agency: identity, permissions, a budget and infrastructure that allows it to transact within defined rules. Eventually, agents may need access to credit, insurance and financial assets.

AI agents are software. Software wants APIs, programmable rules, 24/7 availability and infrastructure capable of operating continuously at machine speed. That makes stablecoins, programmable money and onchain financial infrastructure a natural fit.

I believe Ethereum will become one of the core financial rails for the agentic economy.

Not because AI agents care about crypto. They don't. They care that money can be programmable, composable and available to software without banking hours or a human clicking approve every time.

Traditional banks and payment networks aren't going away. The interesting architecture will connect them with stablecoins and Ethereum.

The future of finance isn't TradFi or crypto. It's increasingly both, with AI sitting on top.

For CEOs and innovation leaders, I'd ask a slightly uncomfortable question:

If your next $10 million of innovation capital were allocated today, how much would you put into traditional transformation programs versus independent AI-native ventures?

I'm increasingly convinced the balance should be shifting.

Banks, insurers, payment networks, asset managers and large enterprises have an unusual opportunity here. They don't just have to buy the next generation of AI infrastructure. They can build businesses around it.

That distinction matters.

Once an AI agent can independently reason, act and transact, we're no longer just changing software or productivity. We're creating a new class of economic actor.

That requires a financial system built for software, one that can handle identity, permissions, payments, capital and risk at machine speed. I believe that system will increasingly connect banks and payment networks with stablecoins and Ethereum.

AI is creating the economic actors. Agentic Finance will give them the financial infrastructure.

That's where I'm going next: what gets built, who wins and where the capital should go.