In practice, a lot of “autonomous” AI agents still spend like an intern who lost the company card. They can draft the brief, compare suppliers, and recommend the next step. The moment the work costs money (a paid dataset, a model call, an API hit), a human has to stop and wrestle with a checkout built for people.
That stall is the business problem San Francisco startup AIsa says it is built to remove. On October 15, 2026, the company announced that Tether and HIVEN had come in as strategic investors on a seed extension, bringing total funding to more than US$9 million. The checks were not published, and no independent audit of the company’s traction claims came with them.
The wager is plainer than a term sheet. Agent commerce needs a payment rail with a budget and a receipt, and Tether, issuer of the USDT stablecoin, is putting its name on a company building that rail.
Agents can act. They still cannot buy.
AIsa, a San Francisco company that says it was founded in 2025, is building one programmable interface where an AI agent can find external data, models, APIs, and software, then pay for what it actually uses. The October announcement gives this a long name: a Trusted Resource and Transaction Network for Enterprise AI Agents. Call it commercial plumbing. The point is paid resources without a separate vendor contract for every tool, and a finance team that can still shut the tap.
The product is not a smarter chatbot. Enterprise AI agents need a way to reach paid services. A finance lead needs a way to stop that access from turning into shadow spend.
Enterprise AI agents need trusted information, useful tools and a governed way to transact.
Jordan Liu, founder and CEO, said that on October 15. What the announcement adds is his vision, not a second quotation. The constraint operators should sit with: global access only works if the company still owns permissions, spending, and the transaction.
Francis Zhan of Tribe Capital, which co-led the earlier seed round with Alibaba, put the market problem in AIsa’s July 7, 2026 funding release. Ads, subscriptions, manual onboarding, and enterprise sales, he said, “do not map cleanly to agents.” A seat license assumes a human logs in. An agent does not want a seat. It wants a metered call, a budget, and a receipt.
In that same July release, Liu reached for an older map. PayPal grew alongside eBay, and Stripe alongside internet commerce. Agents, in his account, cannot operate through human signup and checkout flows. If AI automation is going to do real work, someone has to rebuild the purchase path for software that buys on its own.
One account, a hard ceiling, and a paper trail
The product pairs discovery with the controls a finance lead would actually ask for: permissions, budgets, spending limits, approval workflows, and audit trails. October’s description also adds source signals (provenance, freshness, and quality), so price is not the only thing a buyer can judge.
A hard ceiling is a requirement. It is not a sticker on the feature list. An agent that loops, or that follows a poisoned prompt, will spend until a rule stops it. A paper trail records what happened. It is not sanctions screening, a known counterparty, or a clear answer on who is liable if the agent pays the wrong party. Cross-border control is not solved by a receipt.
The pattern AIsa sketches is small and frequent. One October scenario is an enterprise research agent buying thousands of microservices at US$0.001 per API call. Another is a Shopify merchant agent paying a supplier in Vietnam, a developer in Brazil, or a creator in Kenya. Those are illustrations, not disclosed customer deployments. They still show the shape of the problem: purchases too granular for a monthly invoice, and too exposed for an open wallet.
One named user shows up in the July 7 release, and only there. AIsa said Impossible Finance uses the network to reach models, data, and APIs through a single interface, and pays for actual usage instead of juggling vendor contracts. That is not the same claim as the October pilots. In October, AIsa said it had started strategic pilots with several leading enterprises in financial intelligence, market research, supply-chain management, and commerce. No names, no contract values, and no word on whether any pilot is in production.
Early traction, as AIsa reported it, was steep and unaudited. From February to June 2026, the company said registered agent users grew 150x and aggregate API calls and transactions grew 200x. It also said it had onboarded more than 50, 000 registered agents without paid marketing. Registered is not the same as running in production at a large company, and the release does not include an independent audit.
A buyer still has to decide whether this is a new channel or a layer on payment work already forming at Circle, Visa, and Stripe. AIsa said in July that it integrates with emerging agent-payment initiatives from all three. That is a single-source company claim, not a joint announcement. Verify it before treating the network as the default rail. An enterprise is unlikely to tear out an existing processor just so an agent can buy a dataset.
A small round, read carefully
The financing math is a floor. It is not a valuation. On July 7, 2026, AIsa said in a GlobeNewswire release that cumulative funding stood at US$6.5 million. That figure is total capital to date, including a seed round co-led by Alibaba and Tribe Capital, with Draper Associates, Sumitomo Corporation, Saison Capital, and others participating. It is not, by itself, the size of that seed round.
On October 15, AIsa said the extension with Tether and HIVEN brought total funding to more than US$9 million. The earlier backers remain on the list, along with others. Comparing the two disclosures, later funding is more than US$2.5 million. Because October only says “more than, ” that gap does not reveal the extension’s exact size, and neither release says whether other money arrived between July and October. Valuation, ownership, and each check were not published.
HIVEN is named as a strategic investor. The releases do not identify the firm, its strategy, or the size of its check.
Tether’s scale does not travel with the announcement. Nothing in the funding releases prices the check, and a familiar investor does not prove the network works.
Watch the settlement language, because it moved. The July release said transactions could settle in fiat or stablecoins. October called stablecoins a natural settlement layer for the agent economy and an efficient, programmable payment solution. It did not say every payment would use USDT, and it did not repeat the fiat option. Anyone weighing this for AI for business should not fuse those two statements into a guarantee either way. No share of volume, and no total payment value, has been disclosed.
AIsa’s stated next steps are a wider global resource network, stronger enterprise governance, and transaction infrastructure built to scale.
The popular story is that agents will simply inherit the corporate card. It breaks on unit economics. Thousands of paid calls, or payments to counterparties in several countries, do not fit a seat subscription or a person clicking confirm. They fit a policy.
Treat agent spend as procurement and the gain is specific: fewer vendor contracts, cost matched to usage, and a purchase finance can reconstruct. Treat it as a shared login and you get a bill made of tiny charges. It arrives immediately, and it adds up badly. AI for sales hits the same wall when an agent starts buying lead enrichment or paying a per-contact fee. The meter has to be visible.
Tether’s name on a seed extension does not prove the network works. More than US$9 million does not prove a winner. It does mark where the hard problem has moved. Enterprise AI is no longer only a contest over which model reasons well. The contest is also over which system is allowed to spend, on what, and who can reconstruct the decision afterward.
Questions worth asking before an agent gets a budget
-
What is AIsa actually selling to enterprises?
A governed way for AI agents to find, use, and pay for outside data, models, APIs, and software through one interface, with usage billing, permissions, spend limits, approvals, and an audit trail. It is not a claim that the agent is smarter than the model it calls.
-
How much capital is in the company, and what is still hidden?
AIsa reported US$6.5 million in cumulative funding on July 7, 2026, and more than US$9 million on October 15, 2026. Valuation, ownership, each investor’s check, and the exact size of the extension were not published.
-
Do agent payments have to settle in a stablecoin?
Not as a stated requirement. July described fiat or stablecoin settlement. October’s language is stablecoin-forward. It does not say USDT is mandatory. Neither release discloses the share of volume or the total payment value.
-
Is this already live with named enterprise customers?
Impossible Finance is the only user AIsa has named, and that was in July, as a usage-based buyer of models, data, and APIs. October’s pilots in financial intelligence, market research, supply-chain management, and commerce are unnamed, with no contract value and no production status.
-
What should a CFO demand before agents start spending?
A budget the agent cannot quietly exceed, an approval path outside that budget, a record of every purchase, and a clear view of who is being paid. A receipt will not screen sanctions, and it will not stop a runaway or prompt-injected agent on its own.