A few weeks ago we put out a call asking who is building in agentic payments across LatAm. The response was enough to convince us this deserves a longer treatment. Here is how we are thinking about the opportunity, and why we believe the region is better positioned to lead it than most people assume.
There is a lending startup in Mexico City that has built something genuinely impressive. Their AI agent ingests a borrower’s open banking data, scores creditworthiness in seconds, generates a personalised loan offer, and routes the decision through compliance checks without a human touching it. The whole thing runs in under three minutes.
Then the borrower accepts. And someone has to log into the bank portal to send the disbursement.
The agent that just automated eight steps creates a ninth that is entirely manual. A junior finance person, a browser tab, a CLABE number copied from a form, and a transfer that settles the next business day. In a product built around instant credit, the money arrives tomorrow.
This is where agentic payments stands today across the region. The automation is real. The rails are not ready for it.
The global conversation about agentic payments tends to assume a Stripe-enabled, ACH-ready world. That is not the world most founders in LatAm are building in.
Mexico has SPEI for interbank transfers and CoDi for QR-based payments, but neither was designed to be called by an agent at two in the morning settling a contractor invoice. Brazil has PIX, which is genuinely world-class as a real-time rail, but the programmatic layer for agents to interact with it cleanly is still being built. Colombia has PSE, which works, mostly, for human-initiated transactions. Argentina has its own set of rails, complicated further by currency controls that add a layer of compliance logic that no agent today handles gracefully.
The result is a region where five major markets each have different rails, different settlement windows, different compliance frameworks, and none of them were designed with machine-initiated transactions in mind. An agent that needs to disburse across borders hits the full stack of correspondent banking friction: intermediary fees, multi-day settlement, and FX spreads that make micro-transaction economics completely unworkable. A payment costing under two dollars domestically can cost twenty or more the moment it crosses a border. In an agentic world where work is routed globally based on capability rather than geography, that cost structure breaks the model entirely.
Five major markets. Five different rails. None of them designed for machines.
Speed and cost are the obvious gaps. Real-time settlement is not a nice-to-have when an agent has completed a task and is blocking the next step on payment confirmation. The rails need to work at API latency, not bank-hours latency.
Dispute resolution is the sleeper issue. Today’s chargeback framework was built around one question: did the authorised human make this transaction? When an agent makes a suboptimal decision on a user’s behalf, that framework breaks. Who holds liability? In LatAm, where consumer protection regulation is still catching up with digital commerce, this ambiguity creates real exposure. It needs to be treated as an architectural problem from day one, not a terms-of-service afterthought. The founders who embed this into their core product will separate from those who bolt it on later.
Identity and counterparty verification is just as hard. A LatAm agent sourcing a vendor is no longer bounded by geography or network in the way a human buyer would be. It might transact with an entity it has never encountered before, across a border, over an API. Verifying that counterparty is legitimate, not a fraud vector, not sanctioned, cannot require a human in the loop. The infrastructure for machine-readable trust signals between agents essentially does not exist yet in the region.
Fraud controls will also break. The models that banks and payment networks across Mexico, Brazil, and Colombia use to flag anomalous behaviour are trained on billions of human transactions. Humans operate in predictable patterns. Agents do not. A disbursement agent settling forty payments in an hour across six states will look like a compromised account on every existing model. Rebuilding the risk stack for machine-initiated transactions is not a small job. For the right founders it is also an enormous opportunity.
We are not backing a single winner-takes-all bet on agentic payment rails. We think the opportunity is more interesting than that, and more specific. These are the areas we are actively looking to fund right now.
This is the wedge that is ready today. The majority of SMBs across LatAm still manually enter invoice data, copy payment details, and chase approvals over WhatsApp. An agent that reads invoices from email, validates the counterparty, stages the payment for one-click approval, and then executes, creates measurable value immediately. The infrastructure built to serve this use case becomes the foundation for fully autonomous agent-to-business settlement as the market matures. We want to back founders who see the AP/AR wedge for what it is: a route into something much larger, not the end goal itself.
SPEI, PIX, PSE, and cross-border stablecoin settlement should be callable from a single API. Today they are not. A founder building the abstraction layer that lets an agent choose the right rail for a given transaction, in a given market, with the right compliance wrapper, is building something that every AI-native financial product in the region will eventually need. This is genuinely hard infrastructure work and we think it will attract a defensible moat. We are looking for teams with deep knowledge of at least two of the major LatAm payment systems who are thinking natively about agent-initiated transaction patterns rather than retrofitting human-facing rails.
Agents need access to funds but businesses need limits, audit trails, and the ability to set rules at the transaction level. This is a product category that barely exists for human corporate spend, let alone for machine spend. We want to back founders building the equivalent of a corporate card programme designed from first principles for agents: just-in-time funding, programmable spending controls, counterparty allowlists, and full observability for finance teams who need to understand what their agents are doing with company money.
Every existing fraud model in the region was trained on human behaviour. As agent-initiated transaction volume grows, those models will produce an unworkable number of false positives. The team that builds a fraud and risk layer calibrated to machine transaction patterns, and that can plug into the existing banking and payment network stack across LatAm, is solving a problem that will be impossible to ignore. We are particularly interested in founders approaching this as a data business, where the transaction signal from many agents compounds into a proprietary risk model that gets better over time.
KYB and KYC were designed for humans onboarding other humans. When an agent needs to verify that the entity it is about to transact with is legitimate, the current tooling is too slow, too manual, and not designed for programmatic consumption. We want to back founders building machine-readable identity and compliance infrastructure for the agentic economy, with a particular focus on cross-border use cases where the counterparty verification problem is most acute.
This is the contrarian case, and we find ourselves making it more often the more time we spend in the region.
LatAm fintech has a long track record of building around weak incumbents rather than waiting for them to move. PIX proved that a well-designed real-time rail can reshape consumer behaviour at a national scale in under three years. Nubank, Clip, and Konfío proved that founders willing to solve hard infrastructure problems from first principles can build businesses of genuine consequence. The same logic applies to agentic payments.
The founders who are building AI-native financial products in Mexico City, São Paulo, and Bogotá today are doing so without the assumption that the rails will be provided for them. That constraint produces better thinking. A team that has to solve SPEI programmability, cross-border stablecoin settlement, and agent identity verification at the same time is building something that would be very difficult to replicate from a US starting point.
The question is not whether this shift happens. It is whether a LatAm company owns the infrastructure layer when it does, or whether the region ends up integrating whatever the US incumbents ship next. We are backing the former.
Building in agentic payments? We want to hear from you.
We are actively looking to back pre-seed founders working on any of the areas above across LatAm. If you are building programmable payment rails, agent spend infrastructure, fraud tooling calibrated to machine transactions, or compliance tooling for the agentic economy, we would love to talk.
Reach out directly. No deck required to start a conversation.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.