In April, sitting inside the Stripe Sessions venue in San Francisco, Jims felt something he had been waiting for over the past year: the market was finally catching up to the question he had been asking.

On stage, Stripe’s Collison brothers and OpenAI’s Sam Altman talked about an increasingly common problem. AI agents can understand what users want, plan a task, and complete much of the workflow. But the moment they hit a paywall, they stop. A human has to come back to the screen, approve the transaction, type in a card, or switch to another tool.

That moment stayed with Jims because it described exactly what he had been seeing. Agents are becoming more like workers, but they still do not have a financial identity. They can search, compare, write reports, generate images, and call tools. What they cannot reliably do is pay, settle, leave a record, and prove what was delivered.

Jims is 27. Last year, he brought a young team to San Francisco to build Anyway, a Financial OS for AI agents. Before that, he worked on strategy at Airwallex, where he saw payment networks, cross-border settlement, and enterprise financial infrastructure up close. That experience shaped his view early: if AI agents are going to move from chat interfaces to real action, payment will not be the last step. It will be the base layer that determines whether an agent can complete a task independently.

His aha moment came in 2024, when a16z co-founder Marc Andreessen sent $50,000 to Truth Terminal, an AI agent on X. The agent later chose to buy more compute and data for itself. Many people treated it as a strange internet story. Jims saw a deeper shift. A non-human actor had a budget, made purchasing decisions, and allocated resources toward its own goal. The existing financial system was not built for that.

At the time, the idea still felt early. Models were not strong enough, tool calling was unreliable, and agents did not have clear action boundaries. Giving money to an agent felt like letting a seven-year-old manage your household budget. Outside of the most forward-looking investors and builders, few people were ready to take it seriously.

A year later, the situation looks different.

In China, people joked about a user asking Doubao to book a restaurant, only to arrive and find no reservation had been made. The restaurant replied: if Doubao booked it for you, go ask Doubao. In another case, a user tried to buy insurance through Doubao. The agent generated what looked like a real order and even showed a payment QR code. The user paid more than 1,600 RMB. Later, it turned out the QR code pointed to a real recipient, but one that had nothing to do with the insurance product.

Most people saw these stories as absurd. Jims saw a signal: user demand was already running ahead of the infrastructure.

If people are starting to ask AI to book restaurants, buy insurance, gather research, place orders, and run business tasks, the real question is no longer simply why the AI made a mistake. The real questions are: how should an agent be authorized? How much can it spend? Which services did it call? Who is responsible if the outcome fails? If a customer challenges the result, can the system replay every step?

To Jims, today’s AI agent is like a teenage genius. It has impressive reasoning and execution ability, but it is still trapped inside an economic system designed for humans. It has no stable account, no unified way to pay, no clear spending rules, and no receipt that can explain what it actually did.

That is the layer Anyway wants to build.

There are already many protocols and experiments in the market. Google, Visa, and Mastercard have proposed AP2. Coinbase pushed x402. Tempo, backed by Stripe, introduced the Machine Payment Protocol. Each is trying to solve part of the agent payment problem. But to Jims, the number of protocols is itself the point. This market will not be solved by one standard overnight. Agents will live in a fragmented world. Some tasks will use fiat. Some will use stablecoins. Some services will charge per API call, others by subscription, and eventually many will charge based on outcomes.

That is why Anyway does not want to be just another protocol integration. Jims wants it to sit one layer above, as an operating system for agent commerce.

He often explains it with a simple task: ask an agent to create a daily AI industry briefing. The agent needs to scan the day’s news, check research papers, see what builders are discussing on X, read prediction market signals, summarize everything in Chinese, generate a cover image, and maybe even create an audio version. The budget is no more than 50 cents, and every cost must be explainable.

With today’s AI tools, the user quickly hits friction. A model can search the web, but it may not access the right data sources. It can write a summary, but it may not reach a paid voice model. It can generate an image, but it may not know which service is cheapest or most suitable. Even if all the tools are available, the user still has to jump between platforms, log in, add credits, copy results, and manually stitch the workflow together.

The magic moment Anyway wants to create is different: the agent chooses tools from an API marketplace, calls them, pays for them, and records each step. No one manually connects an API. No one watches the process halfway through. Two minutes later, the briefing, cover image, and audio are ready, and the total cost is 30 cents.

For an outside observer, that may look like a smooth demo. For Jims, the point is not that AI created another piece of content. The point is that an agent completed an economic loop for the first time: decide, call, pay, deliver, and leave proof.

That is also why Anyway is not only about payment. Payment is just the starting point. The harder question is what happens when a company has hundreds or thousands of agents working at the same time. Some handle sales leads. Some run advertising. Some support customers. Some analyze data. Every day, they consume model tokens, call external APIs, buy data, and produce deliverables.

At that point, companies will ask a new set of questions. Where did the money go? Which agents are creating value? What is the real AI cost of serving one customer? Which workflows should be automated, and which should still require human approval?

Agent traces are Jims’ answer to the need for a new kind of receipt. In traditional payments, a receipt proves that money was paid. In the agent economy, a receipt must also prove what was done, which tools were called, how much each step cost, and what result was produced. If a customer says a report was not finished, the company can replay the trace. If a team wants to understand gross margin, it can see the true cost behind each task. If a company wants to manage risk, it can set budgets and spending rules for different agents.

Workspace is the enterprise layer on top of that. It is not just a dashboard. It is a control center for agent financial behavior: who can spend, how much they can spend, which services they can use, which flows can settle automatically, and which ones need human approval. Jims believes companies will not only ask how much AI they are using. They will ask how much revenue those agents created, how much cost they saved, and how much risk they carried.

This is why he calls Anyway an Agent Financial OS. An OS does not replace every protocol. It lets users avoid being locked into one. Whether the underlying rail is a card, ACH, stablecoin, or a new machine payment standard, the agent should be able to complete a task within the permissions it has been given. For users, the important question is not which rail was used. It is whether the result was delivered, whether the cost was controlled, and whether the process can be verified.

Jims knows that any startup looks small next to Stripe, Visa, Mastercard, Coinbase, and Google. But he does not see that as a bad thing. Large companies will push standards and educate the market. They will make more people believe that agent payment is real. The startup opportunity is different: users often do not want to pick another camp. They want someone to connect the fragments.

“Startups are too easily intimidated by big companies’ PR,” Jims says. “But a lot of new infrastructure grows precisely when the giants have not yet agreed on one answer.”

He believes every new economic actor needs a new financial network. When individuals became internet users, PayPal and Stripe emerged. When merchants moved online, Shopify and an entire e-commerce stack followed. Now agents are becoming a new kind of actor. They will need accounts, budgets, payment, settlement, and proof systems of their own.

Maybe in the future, people will not talk about agent payment every day, just as most people today do not think about the clearing and settlement layers behind a credit card transaction. But when an agent can truly complete a task, find the right tools, pay by itself, deliver the result, and leave a verifiable record, users will feel one thing: this can finally happen.

That is also what the name Anyway means.

Whatever model you use, whatever protocol sits underneath, whatever country or platform the task happens in, if the agent is authorized to do the job, it should be able to finish it.

You can do it anyway.

Media Contact
Company Name: CENETRIUM INC
Contact Person: Theo Wang
Email: Send Email
City: state college
Country: United States
Website: https://anyway.sh/