The AWS–Domo Blueprint for Enterprise Agents: What It Takes to Move From Pilot to Production

Building an AI agent is one thing. Scaling it across your enterprise without breaking trust, compliance, or your data foundation? That's where most organizations get stuck.

In the Domopalooza 2026 breakout session Agentic AI Journey: Your Roadmap from Assistants to Autonomous Platforms, Nick Simha, ISV SA Leader for Data, GenAI, and Emerging Markets at AWS, laid out a practical AWS–Domo blueprint for enterprise agents. His core message cuts through the hype. Scaling agentic AI isn't blocked so much by model costs or technology readiness. It's blocked by data reliability, access control, and governance.

"What we want to tell you is about how AWS and Domo work together to bring agents to life for your organization and how you in this room can make it real when you go back from here," Nick explained.

So, if you're evaluating how to move from agent pilots to production deployments, here are three insights from the session that can shape your roadmap.

Divide the work between foundation and integration

The AWS–Domo blueprint for enterprise agents separates responsibilities into two layers. AWS provides the AI foundation: infrastructure, security, model access through Amazon Bedrock, and production controls via Agent Core. Domo provides the enterprise tool and data integration layer: connectors to over 1,000 data sources, a semantic layer for context, low-code workflows, and governance toolkits.

This division matters because agentic systems ultimately run as code on infrastructure, but they require additional AI capabilities like LLMs, memory, and model choice. By positioning AWS as the secure foundation and Domo as the governed business experience layer, the blueprint offers an easier path to building and deploying agents securely at scale.

"Security is really something that you have to be very, very careful about, especially when you are going into a new field, because if you lose trust, then it takes a long time to rebuild it," Nick noted.

For practitioners, this means you can focus on what you control: your data, your workflows, and your governance rules. You don't need to build the AI plumbing from scratch.

Here's how to think about the division of labor:

• AWS Bedrock: Access over 100 models through a single API, with consistent guardrails as new models arrive
• AWS Agent Core: Handle tenant separation, access control, and production scaling without writing code
• Domo connectors: Pull data from fragmented enterprise systems into a unified environment
• Domo Semantic Layer: Add business context so agents understand what they're working with
• Domo governance: Apply the same access controls and compliance rules you already use for dashboards

Understand the 4 building blocks agents need

Enterprise agent readiness hinges on four components: an LLM to plan and execute, knowledge bases and business rules for compliance, tools (APIs) to take actions, and memory to personalize and persist context. The AWS–Domo partnership aims to deliver all four in an easy-to-use manner.

"You need all of these four things for really the agents to work," Nick emphasized.

What makes this framework useful is its simplicity. You can audit your own readiness by asking four questions:

• LLM access: Do you have a way to access models that can plan and reason? Amazon Bedrock provides access to over 100 models, and Simha noted that models are becoming more specialized. Anthropic's Claude excels at coding, while other models perform better in medical domains.
• Knowledge: Does your agent have access to business rules, policies, and domain knowledge? If you're booking travel, the agent needs to know your corporate policy. If you're doing financial reconciliation, it needs compliance rules.
• Tools: Can your agent call APIs and take actions? This is where the Model Context Protocol (MCP) becomes important. MCP is a way for agents to talk to APIs and data without building custom integration.
• Memory: Can your agent remember context across sessions? Personalization and continuity require both short-term and long-term memory.

"MCP, for those who are not familiar with it, it's a way for agents to talk to APIs and your data easily without having to build custom integration," Nick explained. He highlighted that Domo's MCP server launch means "all the data that you have and everything that you have can be accessed very easily by these agents."

Before you build your first agent, map out which of these four components you already have in place and which need work.

Fix your data foundation before scaling

Data reliability is the number one barrier to scaling agentic AI to production, even when technology and costs are not the blocker. Simha cited that 90 percent of organizations say data reliability is a barrier to scaling generative AI to production, and less than 20 percent feel they have a mature data foundation.

"Really, the data is the problem," as Nick put it.

The risk is that if an agent acts on bad data, it can take bad actions. For example, an inventory management agent working with inaccurate stock levels could trigger unnecessary reorders or miss critical shortages. An agent doing financial reconciliation with unreliable data can have cascading effects across your entire system.

This is where data practitioners have an advantage. You already know how to classify data, identify what's sensitive, control who has access, and govern data lifecycles. Those same skills translate directly to agent governance.

"The same rules for agents as for humans," Nick advised. "If your employee cannot do it, an agent should not be able to do it either."

Here's a practical rollout approach based on his guidance:

• Start with trusted data: Pick a use case where you have high confidence in your data quality and governance
• Avoid customer-facing scenarios initially: Nick advised not starting with a customer-facing use case. "Pick something where the failure is not catastrophic," he said.
• Apply existing governance rules: Use the same PDP (personalized data permissions) and access controls you use for human access
• Build and show, don't just talk: Nick encouraged his listeners, "Go build your first agent. Don't wait. When you show something to your managers...it is more compelling than talking about it"

The AWS–Domo blueprint for enterprise agents gives you a framework, but the real work is getting your data house in order. Model costs have dropped by orders of magnitude. The technology has been proven. What's missing in most organizations is the data reliability and governance foundation that lets agents act autonomously without creating risk.

You already have the skills to solve this problem. The question is whether you'll apply them to agent governance before your competitors do.

Ready to dig deeper into the full roadmap? Watch the complete session from Domopalooza to see the live demo and get more implementation details.