JPMorganChase

Building AI with Purpose from the Start

The experiences that shaped my approach most were not tied to one role. They came from seeing the same pattern across technical, operational and highly regulated environments: the quality of a model matters, but it is never enough on its own. AI creates value only when it is connected to a clear business purpose, clear ownership and clear controls.

Over time, I stopped viewing compliance and risk management as downstream review functions. In mature organizations, they need to be part of the design from the beginning. If governance appears only at the end of the process, the team is already late. This is especially important in areas such as compliance, legal, financial crime and resiliency, where ambiguity can create real operational and reputational risk.

Another shaping experience has been building teams that can move from experimentation to repeatable delivery. Strong intake, documentation, version control, stakeholder alignment and decision rights may sound procedural, but they are what separate innovation theater from enterprise capability. More recently, the rise of generative and agentic AI has reinforced that point. We are no longer only managing model outputs, we are managing workflows, incentives, escalation paths and human accountability.

Making Governance a Driver of Innovation

I do not see innovation and governance as opposing forces. In strong organizations, governance is what makes innovation durable. The wrong approach is to let teams experiment freely and then try to wrap controls around the solution later. The better approach is to define approved patterns up front: what data can be used, what requires human review, what evidence is needed before production, how performance will be monitored and who owns the risk.

In practice, that means moving from broad principles to operating mechanisms. Not every AI use case should be governed the same way. A summarization assistant, a customer personalization model and a financial-crime workflow tool carry different risks and should have different control expectations. Multidisciplinary review also matters because legal, compliance, security, architecture, data and business teams each see different parts of the risk.

“In strong organizations, governance is what makes innovation durable.”

Transparency is equally important, but I define it broadly. It is not only about explaining a model externally. It is also about internal clarity - what the system does, what it is allowed to do, what data shaped its outputs, how exceptions are handled and who is accountable. The goal is to make governance part of how teams build, not an external review ritual.

Creating Teams That Deliver with Confidence

The most important lesson is that teams do not bridge advanced technology and risk management by accident. Leaders should design that bridge deliberately. Highperforming AI teams need more than strong technical talent, they need a shared operating model.

That means clear intake, defined roles, stakeholder commitment, timelines, documentation and quality controls. These practices are not bureaucracy. They are the foundation that allows technical teams to collaborate productively with risk, legal, compliance, operations and business partners.

A second lesson is that language matters. Technical teams often speak in terms of architecture, model performance and experimentation. Risk leaders think in terms of controls, materiality, accountability and residual risk. Business leaders focus on outcomes, adoption and timelines. If these groups are each “right” in their own language but disconnected from one another, projects stall.

The third lesson is that trust is built through evidence, not aspiration. Teams need to show how they test, monitor, document and escalate issues. In regulated environments, credibility comes from producing both innovation and evidence. The leadership challenge is to create a culture where experimentation and rigor are both visible markers of excellence.

Preparing for the Future of AI Governance

AI and machine learning will transform compliance, oversight and risk functions in three major ways. First, they will move these functions from retrospective review toward more continuous and proactive detection. Second, they will help organizations operate at a much greater scale across documents, communications, workflows and other unstructured data. Third, they will force organizations to rethink governance itself, because oversight is shifting from static models to dynamic AI-enabled processes.

In practical terms, I expect AI to become less of a point solution and more of a control layer embedded in operations. In compliance and oversight, that means better triage, prioritization, anomaly detection, document understanding and investigative support. In risk management, it means earlier signal detection, richer decision support and more adaptive monitoring.

The key opportunity is not simply automation. It is augmenting institutional judgment with faster pattern recognition, broader coverage and better evidence. At the same time, the control burden will grow. As systems become more generative and agentic, organizations will need stronger evaluation, monitoring, provenance and failsafe design. The organizations that win will not be the ones that deploy the most AI. They will be the ones that make AI legible, governable and aligned with real operating decisions.

Leading at the Intersection of AI and Risk

My recommendation is to stop thinking in separate tracks. The most valuable professionals in this space will not be purely technical, purely regulatory or purely managerial. They will be translators who can move across those worlds with credibility.

First, build real technical depth. You do not need to be the best model builder in the room forever, but you need enough grounding to understand how systems are trained, evaluated, deployed and monitored. Second, take governance seriously as a craft. Learn how model risk, documentation, testing, controls, data lineage and escalation design fit together. Third, spend time close to operations. Too many AI careers are built around demos and prototypes that never survive contact with real business constraints. Work on systems that have owners, users, deadlines, controls and consequences.

Finally, develop judgment and communication. As agentic AI becomes more prevalent, the hardest problems will involve ambiguity, incentives and governance more than benchmark wins. The professionals who can explain those issues clearly to executives, engineers and control partners will shape the field.