AWS launched Amazon Bedrock AgentCore payments in preview at its Financial Services Symposium in New York, enabling AI agents to autonomously pay for services they consume during task execution. The feature, built in partnership with Coinbase and Stripe, eliminates months of custom billing integration work that previously stalled agent deployments at paid API endpoints, according to SiliconANGLE.
The Production Shift
The symposium surfaced a pattern across banking, insurance, and capital markets: firms are moving agentic AI out of pilot programs and into production workflows that generate revenue or cut costs measurably.
In 2025, 45% of IT decision-makers ranked generative AI as their top budget priority, according to AWS research. By 2026, nearly 100% of financial services respondents said AI budgets would increase or hold steady, with 41% focusing investment on optimizing AI workflows already in production, per SiliconANGLE.
Vanguard: Weeks to One Day
Vanguard rebuilt its inheritance and account transfer workflow on AWS using Amazon Bedrock to automate document scanning and classification. The process, which previously took one to two weeks of manual review during some of the most stressful moments in a client’s life, now completes in a single day. Andy Li, senior manager and head of IT at Vanguard, and Vikram Singh, director of delivery management at EPAM Systems, detailed the implementation at the symposium, per SiliconANGLE.
Agents That Pay for Themselves
The AgentCore payments preview addresses a specific blocker in agent architecture. Until now, an AI agent hitting a paid API endpoint simply stopped. Developers spent months building bespoke billing integrations and managing separate API keys across every service their agent needed. Preethi CN, director of AgentCore at AWS, said the roadmap extends beyond micropayments to commercial transactions like flight and hotel bookings, per SiliconANGLE.
Observability as Prerequisite
Banco Itaú Unibanco, Brazil’s largest private-sector bank, cut incident volume by nearly half and reduced resolution time by more than 30% after consolidating from over a dozen siloed monitoring tools onto Datadog’s unified platform. John Trapani, Datadog’s field CTO for financial services, told SiliconANGLE that consolidated observability is increasingly a prerequisite for deploying AI capabilities like agentic SRE.
Digital Market Twins
Nvidia’s Ioana Boier, global head of capital markets, described how Nasdaq is using generative AI to build digital replicas of limit order books, effectively market twins that trading firms can stress-test strategies against. Boier told SiliconANGLE that the shift changes the economics of quantitative research: “It used to be that quantitative research was about making certain assumptions or looking at the historical information. That’s never repeating itself. The past is not repeating itself.”
Why Financial Services Moved First
The symposium’s throughline: financial services has clearer ROI signals for agent deployment than most verticals. Document processing, compliance monitoring, and transaction execution are high-volume, rule-heavy workflows where autonomous agents can demonstrate measurable cost savings within quarters, not years. With AgentCore payments removing the billing integration barrier, the next test is whether agents autonomously spending money can be governed with the same rigor financial institutions apply to human employees spending money.