Harshal Bhangale

Harshal Bhangale works on the practical problem of letting AI agents pay for the resources they need to complete tasks. In “Why Your AI Agent Needs a Wallet”, he introduced himself as an engineer on Circle’s agentic product team. His argument starts with a familiar interruption: an agent can research a question and choose a useful service, then stop because access requires a purchase, account registration, or API credentials.

Building for frequent, tiny purchases

Circle identified him as a staff software engineer in its March 2026 technical article on Circle Nanopayments, which he co-authored with fellow Circle engineer Bhushit Agarwal. The article explains how offchain payment processing and batched blockchain settlement support USDC transfers as small as $0.000001. These mechanisms address a different purchasing pattern from conventional checkout: agents making frequent, tiny payments for individual resources.

Bhangale argues that human-oriented purchasing flows fit this pattern poorly. An agent may need one piece of data, an inference request, or a brief use of a communication service, without subscribing to an entire product. Requiring a person to register an account or approve each purchase interrupts the workflow; transaction fees can also make very small purchases uneconomical. He presents payments that are immediate, inexpensive, programmable, and continuously available as infrastructure for completing these tasks.

A trip-planning test with two agents

His World Cup trip-planning demonstration gives the argument a concrete test. Two Claude Code sessions receive the same assignment: research flights, hotels, tickets, and stadium logistics, then send an email and deliver a phone briefing. The session with a funded Circle wallet pays providers for services, sends the email, and places a call that answers a follow-up question about reaching the stadium. The other session creates a Gmail draft and reports its findings in the terminal because it lacks the capabilities needed to complete the communications. The demonstration shows how paid services expanded that session’s capabilities; its outcome depends on the tools available to each session.

How agents buy access within a budget

• Spending limits: According to Bhangale, wallet-enforced spending limits let the funded agent make individual purchases within a person’s budget. Bhangale demonstrates a 15-cent maximum for a request and describes session and daily caps. He describes placing these controls in the wallet as allowing frequent small purchases without requiring human approval for each one, while retaining limits on how much the agent can spend.
• Payments in API requests: He also describes payments inside API requests: a server responds with payment requirements, the agent signs a wallet authorization, and it retries the request with payment. This makes buying access part of using a resource, rather than a separate checkout process.
• Offchain authorization and batched settlement: Stablecoins alone do not remove the cost and latency of settling every purchase on a blockchain. Bhangale explains that gas fees can consume a substantial share of a tiny payment, while demand for shared blockchain capacity can delay access. Circle describes Nanopayments as using offchain authorization and batched settlement to separate those steps. After funds are deposited into a smart contract, the agent signs an authorization specifying the recipient and amount. Bhangale says the merchant relays it to Circle and receives confirmation that funds are available within a few hundred milliseconds, allowing it to release the resource before the payments are settled in an onchain batch.

Bhangale presents these components as parts of the Circle Agent Stack: wallets hold funds and enforce spending limits, merchant SDKs add paid access to endpoints, and USDC with nanopayments supplies the payment infrastructure. His focus is the connection between purchasing and execution—giving agents a controlled way to obtain resources and finish the work requested of them.