Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

AI will not create leverage until leaders know where judgment belongs

Leaders must redesign approvals, exceptions, and decision rights around the work.

[Photo: Getty Images]

Profile photo of Rishi Katdare

Sr. Leader, Networking, Edge & Perimeter Security Services. · Amazon Web Services

Years before AI agents became an executive priority, I saw an operating problem in enterprise software.

My team had automated customer access to software as the business moved from one-time licenses to subscriptions and usage-based plans, and we moved routine plan changes into the product so customers could complete them without a salesperson.

The operating gain came from identifying routine choices the workflow could handle and exceptions that still required human judgment.

That distinction matters even more with AI. A system can summarize, recommend, route, and increasingly act, but if leaders leave the old approvals, reviews, and escalation paths intact, the organization gets better output without a better operating model.

That is where AI often disappoints inside the enterprise. The task gets faster while the decision path stays untouched. Summaries arrive sooner. Recommendations improve. Then the same output enters the same meeting, approval chain, and escalation queue. The company pays for the new capability and preserves the old operating burden.

As systems gain more authority, leaders must decide what the system is allowed to do, when it must stop, and who owns the exception.

The result is an expensive middle state. The company pays for the old process, the new system, and the people checking the machine. Output rises, but the operating model gets heavier. That is hesitation with better tools.

Redesign the decision path

With AI agents, the issue becomes sharper. Identity, permissions, and approval boundaries determine whether an agent can merely recommend an action or actually take one. The consequential judgment is increasingly about authority alongside output quality.

In a financial-services workflow, a useful guardrail checks an agent’s authority before an action proceeds. It stops an out-of-scope request, redirects the work to an approved path or requires human review before the agent continues. Human judgment sits at the boundary and the exception.

The licensing work made the same distinction concrete. We removed sales involvement from routine plan changes once the rules were built into the product. A salesperson no longer had to validate a routine change simply because that was how the old process worked.

Human judgment still mattered when a request changed customer rights, commercial terms, or the economics of the transaction.

Keep only controls that still decide something

Reviews, approvals, sign-offs, and checkpoints can all be legitimate controls. They can also outlive the risk they were created to manage. For each review, leaders should name the decision a person makes and the risk that review controls.

The same principle applied to orders placed through equipment partners and sales channels. Routine software access could move through the automated path, while requests that changed customer rights or commercial terms left it for review. Routine requests followed established rules; consequential exceptions reached a person with decision authority.

AI exposes the difference between a control that still protects the business and a ceremony that now protects the process. When the old review survives after its decision value has disappeared, the organization becomes more layered without improving control.

When leaders leave those reviews in place, managers spend time confirming routine work while consequential exceptions wait. Teams build parallel checks because nobody has decided what the new workflow is trusted to do. AI usage rises as the operating model gets heavier.

Leadership determines whether better tools translate into a better decision path.

Use exceptions to test the operating model

Routine cases make automation look good. Exceptions show whether the company knows where decisions belong.

A practical test is to map the decision path around the task. Which cases are routine enough to handle under predefined rules? What condition forces an exception? Who owns that exception? What review disappears once the new path is trusted?

If no meeting, approval, handoff, or review can be retired, the company may have automated output without redesigning the work around it.

Successful implementation is visible in what disappears. Routine cases move without ceremonial approval. Exceptions reach a named owner. Reviews narrow or end. Decision time falls without weakening control.

What success can look like

The scorecard should include work retired, exception quality, and tool use. Adoption measures tool use. Operating leverage shows up in how work moves, how decisions are made, and where accountability sits.

Some decisions should remain human because the stakes, context, or accountability are too consequential to hand off. Human judgment belongs where it changes the outcome.

Before approving the next AI initiative, leaders should answer one question: Where does human judgment move if this succeeds?

A credible answer names the changes to the work, the control path, and the accountability around it. It specifies what stops, what escalates, and who owns the call. That is where AI begins to create leverage.

Rishi Katdare is a senior technology executive at Amazon Web Services.