A former Google software engineer who spent years building the company's ad systems is now warning that the so-called agentic economy — where AI agents buy, sell, and negotiate on behalf of humans — won't work without a fundamental rethinking of how payments settle. The engineer, who worked on recommendation engines, tracking pipelines, and conversion funnels at Google, said the current payment infrastructure wasn't designed for machines that transact autonomously.
Why the engineer started paying attention
Around 2024, the engineer noticed a shift. AI agents were starting to handle tasks like booking travel, ordering supplies, and even negotiating contracts. But the payment rails behind those transactions were still built for humans — slow, batch-processed, and reliant on credit cards or bank transfers. The engineer argues that agents need a native settlement layer, something like a real-time, programmable ledger that can handle microtransactions, conditional payments, and automated dispute resolution without human intervention.
The gap in today's payment systems
Traditional payment networks charge fees that eat into tiny agent-to-agent payments. They also lack the ability to execute complex logic — like paying only if a service is delivered, or splitting a bill among dozens of agents. The engineer's argument echoes a growing concern among developers building autonomous systems: that the financial plumbing is becoming the bottleneck.
Lessons from building ad tech at Google
The engineer's background in ad systems gave a front-row seat to a previous transformation. Ad tech had to evolve from simple banner buys to real-time auctions, fraud detection, and multi-party attribution across dozens of platforms. The engineer sees parallels in agentic payments: the industry will need to build a new layer that handles identity, audit trails, and instant settlement — all without a human approving each step.
What a native settlement layer might look like
While the engineer didn't propose a specific technology, the argument points toward blockchain-like ledgers, central bank digital currencies, or new payment protocols designed for machine-to-machine transactions. The key requirement is trust without intermediaries — or at least trust that can be verified programmatically. The engineer noted that without such a layer, agentic systems will either grind to a halt or be forced to rely on slow, expensive human-mediated payments.
The question now is whether existing payment networks — Visa, Mastercard, ACH, or even the Fed's new instant payment system — can adapt fast enough, or whether a whole new infrastructure will emerge. The engineer's argument suggests the clock is already ticking.
