Q&A: How Polygon’s Becoming the Payment Rail for AI Agents

Built by humans for humans, the global payments infrastructure relies on a matrix of bank accounts, credit cards and manual oversight.

The system runs on one strict prerequisite: you have to be human to play. 

Today, this legacy infrastructure is facing its ultimate mismatch as autonomous AI agents evolve into corporate powerhouses, handling everything on autopilot.

However, they are running headfirst into a banking system that simply doesn't recognise them. 

Tech and payment giants like Google and PayPal already agree that legacy financial rails are structurally incompatible.

Executives believe the next wave of internet commerce to run on crypto rails as AI agents structurally cannot use traditional financial accounts.

In this exclusive Q&A with Technology Magazine, Marc Boiron, CEO at Polygon Labs, explores how Polygon solves this bottleneck by allowing AI agents to seamlessly transact via stablecoins without ever needing a traditional bank account. 

Since AI agents cannot open bank accounts, how is Polygon providing a viable on-chain identity that satisfies regulatory requirements?

This is the foundational question for the agentic economy and the short answer is that AI agents don't need bank accounts.

They need easy access to wallets and verifiable, programmable onchain identity and that's something blockchains do better than the banking system ever could.

On Polygon, agents operate through smart accounts with account abstraction, so authorisation logic, multi-sig, policy checks, rate limits and allow lists are enforced at the wallet layer itself. 

For identity, we're building around ERC-8004, the emerging Ethereum standard for onchain agent identity and reputation, co-authored by MetaMask, the Ethereum Foundation, Google and Coinbase. 

Our Polygon Agent CLI lets a developer register an agent's identity in a single command, and from there, that agent can accumulate reputation, become discoverable by other agents and participate in a trustless economy.

The KYC ultimately sits on the principal, the company or institution deploying or authorising the agent, rather than on the agent itself.

That maps cleanly to how regulators already treat automated systems acting on behalf of a legal person. 

What Polygon adds is cryptographic proof of authority, programmable permissions and a full audit trail at the wallet and identity layer.

In practice, these agents can operate alongside existing banking infrastructure, treasury systems or the Open Money Stack, rather than requiring a completely separate financial system.

How does Polygon enable AI agents to transact with traditional merchants that don't natively support crypto?

Stablecoins on Polygon are the bridge between agentic commerce and the legacy merchant base.

The insight is that traditional merchants don't need to integrate crypto. They need to get paid in dollars, on time, with minimal infrastructure change on their end.

That's a solved problem on Polygon. Stablecoin volume on the network already moves at fractions of a cent per transaction with seconds to finality and the on- and off-ramp infrastructure built around the chain converts stablecoin balances to local fiat at the point of settlement across most major markets. 

The merchant doesn't need to hold a stablecoin, run a node, or learn what a wallet is.

They get dollars, or their local equivalent, in the account they already use which is analysed by analysing the payment intent to the agent and accepted payment methods by the merchant for settlement.

The blockchain is invisible to them, which is the right outcome.

When you combine that with agent-native authorisation at the wallet layer, you get the building blocks of machine-to-machine commerce that actually closes the loop with the existing merchant base.

The agent transacts in stablecoin on Polygon. The merchant settles in fiat. 

Neither side has to compromise on the rails they already trust.

Gartner expects roughly 15% of global retail transactions to be agent-decided by 2028 and the rails for that volume have to look like this if they're going to scale: invisible to the off-chain world, programmable on the on-chain side.

We've collapsed the integration problem. That's the whole game.

With competition from Base and Solana, what specific Polygon features are currently driving 8 million daily agentic transactions on your network?

First, cost and speed.

Polygon is currently processing around eight million transactions a day at roughly US$0.015 per transaction, with 2-3 second finality. When an agent is making hundreds or thousands of micro-decisions a day, fractions of a cent matter enormously at scale.

Second, purpose-built agent infrastructure.

The Polygon Agent CLI is an end-to-end toolkit, wallets, payments, identity and bridging, in a single install, with session-scoped smart contract wallets, per-token spending caps, contract allow lists and time-gated key expiry. 

Crucially, the private key never enters the agent's context.

We host our own x402 facilitator as core infrastructure. Combined with our recent Rio payments upgrade, that gives developers an agent-ready stack out of the box.

We have also doubled down by allocating a portion of transaction fees to accelerate agentic payments which means service providers have optimal margins even at microtransaction scale.

Third, integration into the Open Money Stack. Polygon owns the full payment infrastructure end to end necessary for agents to transact at scale.

This means we are able to have not only the most competitive cost structure, but the most seamless experience when deploying agents to use with a variety of financial primitives such as ramps, cross-chain, liquidity infrastructure, applications and more.

In comparison to other providers, developers have to piecemeal solutions and negotiate contracts with multiple parties.

What are the primary ‘traditional rails’ that will become obsolete first as AI agents move from experimental use to mainstream B2B adoption?

Traditional rails won’t disappear overnight, but the first layers to become invisible are the ones built around human coordination and manual reconciliation.

AI agents operate continuously, transact programmatically and require deterministic settlement, which exposes the inefficiencies of systems designed for batch processing and human approval workflows.

The earliest pressure points are likely to be cross-border payments, invoice reconciliation, treasury operations and fragmented B2B onboarding workflows.

Today, those systems rely heavily on intermediaries, delayed settlement windows, manual compliance checks and siloed databases. 

For an AI agent executing transactions autonomously, those delays become operational bottlenecks rather than administrative inconveniences.

If the next decade of commerce is defined by agentic transactions, what is the biggest hurdle, technical or regulatory, preventing immediate mass adoption?

Regulatory, not technical.

The technology is ready. Agent identity, programmable money and low-cost settlement are all production-grade today. What's missing is legal clarity on liability.

When an agent executes a contract or moves money, who is responsible if something goes wrong: the developer, the deploying company, the end user, or the model provider?

The good news: this isn't a years-long blocker.

Singapore's IMDA published the world's first agentic-AI governance framework in January 2026, including standardised ‘Agent Identity Cards’ that specify capabilities, authorised action domains and escalation protocols and a clear operator-deployer liability split.

The EU AI Act takes full effect in August 2026, though it was negotiated before agentic systems matured, so we'll see updates.

The UAE is taking an innovation-first posture designed to attract agent builders. 

My expectation is that within 12 to 18 months the basic legal scaffolding is in place across the major jurisdictions and at that point adoption moves from experimental to mainstream very quickly.