Amazon’s Q2 2026 earnings call showed AWS accelerating as enterprises move from testing generative AI to building and running production agents.
AWS revenue grew 36.7% year over year, reaching a $169 billion annualized revenue run rate. Amazon also said its AI revenue run rate has passed $25 billion and is growing at triple-digit percentages.
For CX leaders, the bigger signal sits underneath those numbers. Amazon is positioning the infrastructure behind AI, including inference, memory, identity, security, and data connections, as the real route to deploying agents at scale.
That changes the conversation. The next wave of CX automation will depend less on whether a provider can launch a polished chatbot and more on whether it can support reliable, economical workflows across customer and employee journeys. Andy Jassy, President and CEO at Amazon, framed the issue as a production challenge rather than a model race:
“Even after you’ve built an agent, you have a lot of muck to worry about, a production agent needs somewhere secure to run, memory so it holds context, an identity so it can act on a user’s behalf, tools and data to connect to, and a way to watch what it’s doing once real traffic hits.”
That is a useful reality check for CX teams. An agent that can answer a question in a demo is one thing. An agent that can identify a customer, retrieve the right policy, trigger a back-office process, record the outcome, and escalate safely when needed is a different proposition.
Amazon Is Building the Infrastructure Layer for CX Agents
Amazon’s answer is Bedrock AgentCore, a managed set of services for building and operating agents. Jassy said the company developed the platform because connecting the components required for production agents “reliably is hard,” and has stalled many deployments.
The CX implications are immediate. Enterprises want agents that can do more than summarize a conversation or suggest a reply. They want systems that can act across the workflows that shape a customer journey.
That could include an agent that checks an order status, amends a delivery, updates a CRM record, flags potential fraud, or initiates a retention offer. But those actions require access to customer data, business systems, identity controls, and audit trails.
Amazon Connect is part of that broader strategy. Jassy identified Amazon’s contact center service as one of its fast-growing agentic offerings, while Amazon Quick has added autonomous agents that can carry out multi-step work in the background.
For contact center leaders, this points to a more demanding standard for AI procurement. The question is shifting from ‘does this tool have AI?’ to ‘can it safely orchestrate work across the systems our teams already use?’
That shift may favor platforms with deep infrastructure, data, and security capabilities. It may also make fragmented AI point solutions harder to justify when an enterprise needs consistent controls across channels.
Inference Economics Will Shape the AI Roadmap
Amazon’s call also made a clear case that AI economics will increasingly matter in customer experience.
Jassy said customers want inference to sit close to their applications and data. He also argued that agent tool use, reinforcement learning, and post-training workloads run primarily on CPUs rather than AI accelerators.
That is why Amazon keeps emphasizing Graviton, its custom CPU family. Jassy said it can offer 30-40% better price-performance than other options, a claim that matters when CX teams begin running high volumes of agent interactions every day.
Amazon Bedrock is positioned as the company’s cost-effective inference service. The promise is straightforward: enterprises can choose models, connect them to proprietary data, and run them without carrying the full operational burden themselves.
However, CX leaders should remain cautious about the assumption that generative AI will automatically reduce service costs. In a recent CX Today interview, Patrick Quinlan, Senior Director Analyst at Gartner, warned that many organizations are underestimating the operational costs involved:
“I think organizations who expect big cost savings from GenAI will be disappointed.”
Quinlan’s point supports the underlying signal from Amazon’s call. Agentic AI can create value, but it also adds new costs tied to usage, compute, governance, integration, and specialist skills.
CX teams should therefore track cost per resolution alongside more familiar measures, such as containment rate, average handle time, customer satisfaction, and revenue conversion. A cheaper model is not always the lower-cost customer experience if it generates poor outcomes or pushes complex work back to human agents.
Security Will Decide Whether Agents Earn Trust
Amazon also used the earnings call to highlight security as a central enterprise AI concern.
The company recently launched AWS Continuum, which uses agents and business context to discover, prioritize, validate, and remediate code vulnerabilities. Jassy said security now comes up in almost every enterprise AI discussion. Jassy highlighted the scale of that concern:
“It’s hard to talk to enterprises about AI right now without their mentioning security, and we expect Continuum to grow quickly.”
CX leaders have good reason to pay attention. Customer-facing agents need access to sensitive records, policies, payment processes, and internal knowledge. That creates a much larger trust challenge than a standalone chatbot answering basic FAQs.
The strongest deployments will give agents enough context to resolve real problems while limiting what they can see, what they can do, and when they must hand work to a human.
Amazon’s Q2 earnings show that the battle for AI leadership is moving into this operational layer. The winners will be the platforms that make agents useful in real workflows, secure enough for customer data, and efficient enough to scale.
For CX leaders whilst the opportunity is real, so is the discipline required to capture it. That means building around trusted data, clear workflow ownership, and measurable unit economics.
Join the conversation: Join our LinkedIn community (40,000+ members): https://www.linkedin.com/groups/1951190/
Get the weekly rundown: Subscribe to our newsletter: http://cxtoday.com/sign-up
