Agents, Governance, and the Discipline Behind AI That Actually Ships

Getting AI into production is only part of the challenge. Once systems begin interacting with business processes and taking action, organizations must decide how to delegate authority and maintain accountability.

In part two of our conversation with Luise Freese, developer, consultant, and author of “From Prototypes to Production: A 90-Day Playbook for Shipping Enterprise AI”, the discussion moves beyond getting AI into production and focuses on a new challenge: operating AI systems that can make decisions, interact with business processes, and take action. As organizations deploy agents across business processes, the conversation shifts from model capability to something more fundamental: authority, accountability, and governance.

Editor’s Note: To catch up on part 1 of our conversation with Luise, where we discussed why so many AI initiatives never reach this stage, check out: Escaping Proof-of-Concept Prison: The Path from AI Pilot to Production.

The Real Question Is Not Intelligence. It's Authority.

Organizations are increasingly building AI solutions that do more than generate answers. Agents can retrieve information, interact with business systems, and trigger actions. As those responsibilities expand, leaders need to think beyond capability alone. The critical questions become:

What can the system decide?
What can it change?
What can it trigger?
When must it stop?

As Luise puts it, “The boundary is the product. And without this boundary, you do not have an agent, so you only have a very unmanaged risk surface.”

That idea anchors the entire conversation. An AI agent is not defined by how sophisticated it appears, but by the boundaries governing its behavior. She describes an IT support agent that can classify requests, suggest priorities, retrieve approved knowledge, and prepare response drafts. However, when confidence is low, sensitive information is involved, or an action carries financial consequences, the system should escalate rather than act independently. The challenge is not determining what AI can do but defining what AI should be allowed to do.

Governance Separates Agents from Risk

As organizations grant AI agents access to more systems, the consequences of poorly defined authority become more significant. A system that can access procurement platforms, customer records, or operational environments can create both positive and negative business outcomes.

In some scenarios and processes, like new employee onboarding, automating hardware provisioning makes sense. In others, it introduces unnecessary cost and risk. An agent that interprets a vague support ticket such as “my computer isn't working” and automatically orders a new laptop could trigger purchases before troubleshooting occurs, creating avoidable expenses and bypassing established support processes. The difference is not the technology itself; rather, it is whether the organization established appropriate boundaries from the start.

Governance Must Be Built into Delivery

As agents gain access to core business systems, governance shifts from documentation to execution. It becomes the mechanism for defining, monitoring, and enforcing delegated authority in practice.

Rather than treating governance as a review step that happens after development, she advocates building governance into every stage of an AI system's lifecycle. Teams move faster when those guardrails are established upfront because they know how they are expected to operate.

What It Takes to Ship AI Responsibly

Throughout the discussion, Luise outlines several practices to help organizations move from isolated AI successes to repeatable business capability.

Assign Clear Authority to AI Systems

Before deploying an agent, organizations need to determine what decisions it can make and what actions it can initiate. Every delegated action carries consequences, whether it involves customer interactions, financial processes, or internal operations. Clear decision rights help organizations reduce ambiguity and establish ownership when outcomes affect the business. Rather than focusing exclusively on intelligence, leaders should define escalation points and responsibilities before deployment begins.

Establish Boundaries that Support Scale

Boundaries are not an obstacle to innovation. They are what make responsible scale possible. As AI agents gain access to systems and workflows, organizations need clear limits on which actions can be automated, require additional review, and should never occur without human approval. These guardrails create consistency across different use cases while reducing the likelihood of unintended outcomes. As Luise argues, the boundary itself is part of the product because it defines how the system behaves in practice.

Integrate Governance into Delivery

Governance works best when it becomes part of the way solutions are built and deployed. Risk tiers, approvals, classifications, monitoring requirements, and escalation paths should not be bolted on after launch. By incorporating them from the beginning, organizations create delivery processes that are easier to repeat, easier to audit, and easier to scale. Good governance feels less like a barrier and more like a clearly marked path that helps teams move forward with confidence.

Align AI with Operational Practices

AI does not replace established engineering disciplines. Version control, testing, deployment pipelines, rollback procedures, monitoring, access controls, and ownership remain essential. Luise notes that highly visible failures involving AI systems are often preventable when organizations apply the same rigor they have used in software delivery for years. Technical repeatability and organizational accountability must develop together because neither can succeed on their own.

Create a Repeatable Path to Deployment

One successful implementation does not create organizational capability. To bridge that gap, Luise advocates establishing a cross-functional tiger team responsible for turning promising initiatives into functional solutions. Combining expertise in data, software, DevOps, and user needs, these teams focus on building a repeatable path to production rather than simply delivering a single project. The goal is not a single successful deployment, but a repeatable model that future initiatives can follow.

Make AI a Standard Capability

One of the discussion's most practical ideas is that mature AI should eventually become boring. Luise shares that organizations regularly hold innovation weeks, AI hackathons, and experimentation programs because they are still developing the muscle required to operate AI consistently. However, she believes that long-term success comes when AI becomes an ordinary business capability rather than a special initiative. Like finance, infrastructure, or operations, it becomes something teams simply know how to do. Once that happens, AI can move from experimentation to sustained business value.

The Discipline Behind AI that Endures

Successful organizations are distinguished not by the sophistication of their AI systems, but by their ability to govern, operate, and scale them over time.

That requires more than enthusiasm. It requires ownership, governance, engineering discipline, and sustained investment. Leaders must decide whether AI is a side project or a long-term investment worth building in the organization. As Luise notes, organizations that treat AI like a pet project should expect pet-project results. Organizations that invest in the people, processes, and foundations required for production are the ones most likely to realize long-term business value.

The future of enterprise AI may involve increasingly capable models and increasingly autonomous agents, but the lesson remains the same: Intelligence alone is not what makes AI work. Businesses realize value when they define authority, build accountability, and establish the operational discipline required to make AI part of everyday business.

Soundtrack of Shift

Luise chose Bob Dylan’s “The Times They Are a-Changin’. To her, it isn’t a song about comfortable, neatly planned change. It’s about the moment the old order can no longer pretend things will stay the same — which is precisely where enterprise AI catches so many organizations still managing new technology with old structures, governance habits, and success metrics.

Experimentation isn’t the problem. Experimentation without a path to production is. Leaders should stop asking how many AI initiatives they have and start asking how many can survive contact with real workflows, real data, real users, and real accountability. That’s the number that matters.

Episode Resources

#shifthappens Research: 2026 State of AI

#shifthappens Insights:

#shifthappens Podcasts:

Dux Raymond Sy on LinkedIn

Luise Freese on LinkedIn

From Prototypes to Production: A 90-Day Playbook for Shipping Enterprise AI

#shifthappens podcast

Relevant Episodes

Episode 135: Agents, Governance, and the Discipline Behind AI That Actually Ships