Thrilled to introduce nekuda 🥳
So who are we? Let’s start with a few words about us.
We began our journey by building LangChain agents and trying to get them to pay for items online. We quickly realized that while the potential is huge, using existing payment rails is tough. How tough? Well, for starters, our card got blocked. We also found out the hard way that agents can easily go off rails and act too fast or buy things we never intended. It became clear to us that even though the opportunity for agentic commerce is massive, today’s infrastructure simply can’t unlock it.
Our thesis is pretty simple:
Agent-driven transactions will generate significant commerce volumes across the internet.
Current payment rails create friction for agent-driven transactions.
Our mission is to build new rails that support agents paying on behalf of users.
What are the problems with agents paying?
Let’s start from first principles—agents are software. But they have a unique property: they’re non-deterministic. That means you know exactly what input you’re providing (prompt, instructions, etc.), but the outcome is fuzzy and can vary each time.
When we give an agent a means of payment, a useful mental model is delegation: we’re sending software out to purchase something on our behalf. We can instruct agents in many ways: “go ahead and book,” “go ahead and buy,” or “close this for me,” all of which should ideally mean the same thing. The agent needs to interpret our intent and translate it into a deterministic buy action.
In the gap between our intent and the agent’s actual actions, there are two main problems:
Ambiguity: We’re often less clear than we think we are. This ambiguity will be amplified significantly at scale in agentic commerce.
Fraud Risk: Fraud becomes a bigger issue than it is today. Users might dispute transactions by saying, “I didn’t buy it—the agent did,” or agents could be tricked into making unauthorized purchases. We can also expect new kinds of dark patterns to emerge specifically targeting agents in the future.
Many mistakes → low trust → agentic commerce can’t succeed!
So, we need a systematic approach to delegate payments to agents safely. We call this the authorization problem. Our solution: capture the user’s mandate explicitly within every agentic transaction, record it securely, and share it transparently with relevant stakeholders (such as card networks). Our SDK solves exactly this and helps agent developers make sure that their agentic payment system is secure for their users.
We envision agentic commerce developing in three phases:
Discovery focused (Current State): AI primarily helps find products, but humans still finalize transactions, reflecting the traditional e-commerce experience we’ve known for decades.
Autonomous checkout (Emerging Now): Browser agents take the next step, handling the checkout process autonomously. This reduces human intervention significantly, creating a new “human not present” shopping scenario alongside the familiar “card not present” situation.
Fully autonomous commerce (Future Vision): Both consumer-side and merchant-side agents independently negotiate and transact without any human oversight, streamlining commerce to its fullest extent.
Currently, we see three distinct types of commerce agents emerging:
Third-party agents: Platforms that aggregate and simplify online experiences (e.g., Perplexity).
Browser-based agents: Agents manipulating websites directly, automating human-like interactions (e.g., OpenAI Operator).
Platform agents: Created by major merchants or platforms, directly facilitating transactions (e.g., Amazon’s Alexa+).
Our first product directly addresses the second phase: “human not present” transactions. This is an entirely new challenge because the existing payments ecosystem wasn’t built for non-human behavior, significantly increasing errors and transaction declines. Our SDK bridges this gap by capturing explicit user authorization (mandates), securely managing payment credentials, and transparently sharing authorization data with all stakeholders (card networks, issuers, merchants).
Our SDK
We’re currently rolling out our SDK. If you’re building any type of agent with commerce capabilities (e.g. travel, commerce, procurement), we’re confident our tools can help you build agents with embedded secure payment. At the most basic level, we enable your agents to handle users credit cards, manage transactions smoothly, record user-authorization and equip you with best-in-class tools for agentic commerce. Interested? Feel free to reach out directly to founders@nekuda.ai or message us here:
We’re also a flagship partner of the Visa intelligent Commerce program for agentic payments. This is the first card network initiative designed specifically for agents. It provides network tokens—tokens that replace traditional credit card numbers and are already widely used by payment processors and wallets like Apple Pay and Google Pay. With these tokens, agents can securely transact on your behalf, ensuring your authorization is always clear and enforceable. The program is innovative, forward-thinking, and we’re proud to be part of it.
There’s also been significant momentum from leading industry players: PayPal has showcased its stack using MCP , Mastercard introduced its Agent Pay program for agentic payments , Stripe showcased their Order Intent API for agentic payment, OpenAI launched a commerce program allowing payments directly through ChatGPT , and Amazon unveiled the “Buy for Me” beta feature, enabling agent-powered purchases via its Nova model . This momentum marks just the beginning of agentic commerce and payments.
Writing
We believe in writing and openly sharing our thoughts about agentic payments, and we plan to continue doing so right here. We’ve attached a few selected pieces from our past work. If you’d like to stay updated, feel free to subscribe.
— the nekuda team
Let the agents pay!