The payment problem nobody has solved yet
AI agents are completing real, billable work in 2026. They are writing code, running campaigns, processing data, managing workflows. The infrastructure to pay for that work, at the agent level, does not exist yet.
That gap is agentic payments. It is not a niche technical problem. It is the next layer of payment infrastructure that the rise of hybrid human and AI teams is going to demand.
What agentic work looks like today
79% of organisations have adopted AI agents in some form (PwC, 2025). By 2028, 33% of enterprise applications will include agentic AI (Gartner, 2025). These are not chatbots answering support tickets. They are agents completing multi-step tasks, operating autonomously inside workflows, and delivering outputs that have real commercial value.
In a project context, this already happens. A team running a product build might include a human architect, two human engineers, a fractional QA specialist, and an AI coding agent completing specific tasks in parallel. Four of those contributors have a payment rail. One does not.
Why existing payment infrastructure does not work for agents
Current payment infrastructure is built around humans. It requires a name, a bank account, a jurisdiction, and a tax identity. AI agents have none of these by default.
The workaround most teams use today is to roll agent costs into platform subscriptions and call it done. That works at small scale. It breaks when you are running dozens of agents across dozens of projects and need to know, at a project level, what each agent cost, what it delivered, and how to reconcile it against the client invoice.
The accounting problem alone is significant. If an agent completes work that gets billed to a client, that cost needs to sit somewhere in the project ledger. Right now, it mostly does not.
What agentic payment infrastructure looks like
The emerging model is payment primitives exposed as callable tools, specifically via MCP (Model Context Protocol), which allows LLM clients to invoke actions directly inside a conversation.
The primitives are straightforward:
- Create invoice
- Approve payment
- Split across contributors (human and agent)
- Select rail (fiat or stablecoin)
- Settle
- Return settlement status
If a payment orchestration platform exposes these as MCP tools, an AI agent, or a human using an LLM, can trigger the full payment workflow inside a conversation. No app switching. No manual reconciliation. The agent completes the work and the payment follows the output.
Why stablecoin is the natural rail for agent settlement
Stablecoin is not a crypto play here. It is a practical infrastructure choice. B2B stablecoin transfer volumes reached $27.6 trillion in 2024, surpassing Visa and Mastercard combined. B2B stablecoin payments grew 30x in two years, from under $100 million per month in early 2023 to over $3 billion per month by 2025 (Artemis / Castle Island / Dragonfly).
For agent settlement, the properties that matter are: programmability, near-instant settlement, no SWIFT dependency, and compatibility with non-human identities. Stablecoin satisfies all four. Traditional bank rails satisfy none of them cleanly.
Where Petl Pay sits in this
Petl Pay is building the MCP layer. The same orchestration infrastructure that splits and settles payments to human contributors across fiat and stablecoin rails is being extended to handle agent-level settlement. Non-custodial by design, stablecoin-native, and built around the project as the financial unit.
The bet is straightforward. As agentic work becomes standard, every project will include AI contributors that need to be invoiced, reconciled, and settled alongside human ones. The payment layer that handles this natively will sit at the centre of how hybrid teams operate.
What this means for agencies and project teams today
Most teams do not need to do anything differently right now. But the ones building for where work is going are starting to think about agent costs as project-level line items, not just SaaS subscriptions.
The practical question to ask: if an AI agent completes a task on your project tomorrow, where does that cost sit in your project ledger, and how does it get reconciled against the client invoice? If the answer is "it does not," that is the gap agentic payments fills.
FAQ
What are agentic payments?
Agentic payments are the payment infrastructure required to settle, reconcile, and account for work completed by AI agents. As AI agents take on real, billable tasks within projects, the financial layer needs to handle them alongside human contributors, with the same level of invoicing, reconciliation, and audit trail.
Can AI agents have bank accounts?
Not in the traditional sense. Agents can be assigned wallet addresses, particularly stablecoin wallets, which can receive and hold value without requiring a human identity or bank account. This is one reason stablecoin is the natural settlement rail for agent-level payments.
What is MCP and why does it matter for payments?
MCP (Model Context Protocol) allows LLM clients like Claude to invoke external tools and actions directly inside a conversation. If a payment platform exposes its core primitives via MCP, an AI agent or a human using an LLM can trigger invoicing, payment splits, and settlement without leaving the conversation or switching apps.
How soon will agentic payments be standard?
The infrastructure is being built now. Gartner projects that 33% of enterprise applications will include agentic AI by 2028. The payment layer to support this is 12 to 24 months behind that adoption curve. The teams building for it now are the ones who will not be scrambling to retrofit it later.
Is this relevant for small agencies today?
Increasingly yes. If you are using AI tools inside client projects and billing for outputs that include agent-generated work, the reconciliation problem already exists. It just does not have a name yet. Agentic payments is the infrastructure that names and solves it.
