Amazon just put enterprises on notice: the next wave of contact center AI will not be judged by how well it chats, but by how safely it can act.
In Amazon’s Q4 2025 earnings call, Andy Jassy CEO at Amazon zeroed in on autonomous agents as the main path to real AI value, and he argued that the blocker has been governance, not intelligence.
His answer was Bedrock AgentCore, positioned as the control layer that helps agents securely connect to identity, policy, tools, and monitoring, so enterprises can let them touch real backend systems.
During the call, Jassy said:
“Looking ahead, the primary way companies will get value from AI is with agents.”
For CX leaders, that is a meaningful shift. It suggests the market is moving away from 'scripted bots' that collapse when customers go off path, and toward governed autonomy, where agents can pursue outcomes like refunds, account changes, or complex service cases within enforceable guardrails.
The Old Model: Scripted Bots That Break When Customers Go Off Path
For years, contact center AI has been defined by constraint.
In most enterprises, automation was built like a flowchart. You mapped intents, defined branches, and hoped customers stayed on rails. That model worked for narrow, repetitive tasks, but it struggled the moment a customer’s request became messy, multi-step, or policy-heavy.
That is why many “AI in the contact center” deployments have ended up as deflection tools. They answer FAQs, collect basic details, and route interactions. They rarely resolve complex issues end to end.
Amazon Connect has lived in that reality too. It has long used AI to improve routing and self-service, and Amazon has also pushed agent productivity tools, including Amazon Q-style assistance in the broader AWS portfolio. But the ceiling has been governance, not raw model capability.
The New Model: Goals, Tools, and Authority, Not Flowcharts
The emerging agentic era replaces rigid scripts with a different contract.
Instead of telling a bot exactly how to process a refund, you give an agent a goal, the policy to follow, and the tools to act, such as APIs into billing, CRM, fulfilment, and identity. The agent “reasons” through the path, handles edge cases, and completes the work.
That is the promise, and it is also why enterprises have hesitated. When an agent can take action, 'hallucinations' stop being an academic risk and start being a security, compliance, and financial risk.
Jassy described that hesitation directly, and he framed it as the central hurdle between prototypes and production deployments.
Why Bedrock AgentCore Is The Missing Link for Enterprise Autonomy
On the call, Jassy argued that building agents is still harder than it should be, and he said AWS has been building services to make agents possible across models. But his sharper point was what happens next.
Once an agent exists, enterprises still need to connect it safely to the real world, and that is where most projects stall.
“Once agents are built, enterprises are apprehensive about deploying to production because these agents need to securely and scalably connect to compute, data tools, memory, identity, policy governance, performance monitoring, and other elements.”
That is the thesis behind Bedrock AgentCore. It is not positioned as another “bot builder.” It is presented as the governance and connectivity layer that makes autonomy viable inside enterprises.
Jassy did not describe AgentCore as optional plumbing. He described it as a solution to a problem "where a solution has not existed,” and he claimed it is already unlocking deployments.
“This is a new and hard problem where a solution has not existed until we launched Bedrock Agent Corp. Customers are quite excited about Agent Core, and it’s unlocking deployments.”
For CX leaders, the implication is straightforward: the differentiator is moving from scripted conversation to governed action.