By Kortney Rupp, Director, Paragon by Riviera Partners 

AI made the work faster. Too bad the approval chain didn’t get an upgrade. 

In the first article in this series, we looked at why faster work can still get stuck. Here’s how to fix it by deciding what AI can do, where people need to apply judgment, and who has the authority to say, “This is ready.” 

Only 19% of organizations reach Advanced AI execution maturity, as our 2026 Future of Tech Leadership survey shows. Those organizations combine unified technology structures, hands-on leadership, and governance built in from the start. Here are five steps for putting it into practice. 

1. Design the workflow 

While leadership debates where AI belongs in the business, employees aren’t waiting. Some are using it to tidy up emails. Others have quietly rebuilt half their workflow around it. And every few months, the tools get more capable, so the gap between what’s possible and what’s been approved keeps growing. Without clear direction, the line between human and AI work still gets drawn. It’s just drawn by whoever happens to build the process, one team at a time, with no shared standard. 

The fix is to make that line a deliberate leadership decision. Start with one workflow that produces a clear business outcome. Map it from start to finish, including handoffs, approvals, and exceptions. Then decide, step by step, whether AI does the work, AI does it with human review, or a person does it. 

Base each decision on three questions: How reliable does this step need to be? What happens if it goes wrong? How much judgment does it require? From there, set clear points where AI stops and a person steps in. Finally, name one owner for the end result. Not for each step, but for the whole outcome. If everyone owns a piece, no one owns the result. 

2. Build the shared backbone 

Once you’ve defined the workflow, separate the capabilities that benefit from enterprise-wide consistency from the work teams should own locally. You can organize responsibilities into four connected layers. 

Layer What it does What breaks without it 
Central AI hub Sets AI priorities, enterprise architecture standards, and risk guardrails Teams make incompatible choices and pursue competing priorities 
AI platform Provides reusable models, tools, approved agents, data access, integrations, permissions, and evaluation systems teams can readily access Teams duplicate infrastructure and controls, increasing cost and complexity 
Embedded AI pods Redesign workflows within business functions using small, temporary teams Companies deploy tools without changing how the work gets done 
Outcome squads Operate and improve workflows against defined business outcomes using persistent, cross-functional teams Accountability fragments after implementation, leaving no team responsible for the complete result 

Pods should transfer the redesigned workflow to an ongoing owner before moving on, while outcome squads remain accountable for its performance and improvement. These layers are responsibilities, not necessarily a new AI department. The hub sets direction, the platform makes shared capabilities usable, and pods and squads apply them within the business. 

3. Give teams authority to deliver 

Teams can’t own an outcome if they need approval for every meaningful decision. Define which decisions the team can make independently, which require consultation, and which need to be escalated and to whom. Cover priorities, architecture, spending, and data access, and name who can resolve decisions outside the team’s authority. If you can’t list the last three meaningful decisions the team made without escalating, they have authority in name only. 

Build the team around what it takes to deliver the result, including product judgment, engineering expertise, domain knowledge, and access to the shared platform. Then pay attention to where the work keeps stalling while the team waits on someone outside it. That’s your signal. If the team needs that skill regularly, bring it into the team. If it’s a need many teams share, like security or data access, make the central support fast and dependable enough that nobody has to queue for it. 

Waiting can also be a problem inside the team. If only one person understands a critical part of the system, the whole squad slows down whenever that person is busy or away. Strong squads spread enough technical depth across the team that work keeps moving. They also make it safe for anyone to challenge a decision or flag a problem early before it becomes expensive. And the person responsible for judging the output has the authority to make the call, without getting dragged into a wider debate about priorities that stalls everyone else. 

4. Build governance into the work 

In our research, 87% of Advanced organizations bring cybersecurity, legal, and governance into the work during initial design. Among Emerging organizations, 56% bring them in just before deployment or only when something comes up. That timing matters. When governance shows up at the end, the final review can send the team back to rebuild what it just finished. 

Decentralized teams move faster when guardrails are part of how they work from the start, not a checkpoint at the finish line. Set approved architecture patterns, quality thresholds, access permissions, monitoring requirements, and escalation rules up front. Wherever possible, build those controls into the shared platform so teams meet them by default and don’t need to request them. 

Oversight must continue after deployment. Assign responsibility for investigating failures, reviewing changes, and deciding when to pause a workflow. Teams need to know who can make a timely decision when established guidance no longer covers the situation. 

When specialists move into cross-functional teams, their practices can slowly drift apart. Over time, each team’s code, designs, and standards start to look different, and the work becomes harder to evaluate, maintain, or reuse. Engineers, designers, and other specialists still need a home for their discipline. Shared reviews and communities of practice give them one, keeping expertise sharp and standards consistent across the organization. 

5. Keep leaders close to execution 

For years, engineering leadership meant moving further from the work. The bigger your team, the more of your time went to planning, status updates, and coordinating between groups. AI is reversing that. As it takes on more of the information gathering and coordination, the leaders who stand out are the ones who stay close to the work. 

In our research, 42% of Advanced organizations say their technology leaders spend more than half their time directly involved in execution, compared with 18% of Emerging organizations. Staying close to the execution calls for a player-coach approach. These player-coaches set direction and develop people, but they also weigh in on architecture, product decisions, governance, and delivery. They can review AI-generated output and tell whether it’s good. They know which people on their team are using AI well and which are struggling. And they understand what the tools cost, not just what they produce. 

For engineering managers who have spent years leading through layers, this is a real shift. The first step is rebuilding technical fluency: getting hands-on with the tools your team uses and learning how AI is changing the work itself. Without that, it’s hard to judge output, coach people, or make credible calls. 

The organization has to make this role possible. If engineering leaders are rewarded for growing headcount and product leaders for shipping more features, those incentives work against smaller teams owning better outcomes. Review performance measures, compensation, and role scope so they reward the behavior you want. 

Be careful with flattening, too. AI can reduce layers of management, but removing a layer doesn’t remove the coordination it handled. Unless that work has been automated, simplified, or clearly reassigned, it lands on whoever is left. Take the work out first, then move the people. 

What this means for your organization 

An AI-native operating model decides how much of AI’s capacity your business can actually use. It’s the difference between faster work reaching customers sooner and faster work waiting in a longer approval queue. And once the design is right, it changes who you need to hire: fewer coordinators, and more leaders who can direct, judge, and deliver the work themselves. 

Continue the series 

  • Part 3: The Talent Model — who to hire for the builder layer, how to interview for judgment, and how to level people when output stops tracking headcount. 

Frequently asked questions 

What is an AI-native operating model? 

An AI-native operating model decides how much of AI’s capacity a business can actually use. Organizations that reach Advanced AI execution maturity, 19% of those surveyed, combine unified technology structures, hands-on leadership, and governance built in from the start. Putting it into practice takes five steps: design the workflow, build the shared backbone, give teams authority to deliver, build governance into the work, and keep leaders close to execution. 

How do you decide what AI does and what people do? 

Make the line a deliberate leadership decision rather than leaving it to whoever builds the process. Start with one workflow that produces a clear business outcome and map it from start to finish, including handoffs, approvals, and exceptions. Then decide step by step whether AI does the work, AI does it with human review, or a person does it, based on three questions: how reliable the step needs to be, what happens if it goes wrong, and how much judgment it requires. 

What are the four layers of an AI-native operating model? 

The central AI hub sets AI priorities, enterprise architecture standards, and risk guardrails. The AI platform provides reusable models, tools, approved agents, data access, integrations, permissions, and evaluation systems. Embedded AI pods are small, temporary teams that redesign workflows within business functions. Outcome squads are persistent, cross-functional teams that operate and improve those workflows against defined business outcomes. These layers are responsibilities rather than necessarily a new AI department. 

What is an outcome squad? 

An outcome squad is a persistent, cross-functional team that operates and improves workflows against defined business outcomes. Embedded AI pods transfer a redesigned workflow to the squad before moving on, and the squad remains accountable for its performance and improvement. Without that handover, accountability fragments after implementation and no team is responsible for the complete result. 

How do you tell whether a team really has decision-making authority? 

If you cannot list the last three meaningful decisions the team made without escalating, they have authority in name only. Real authority means defining which decisions the team makes independently, which require consultation, and which need escalation and to whom, across priorities, architecture, spending, and data access. 

When should governance be involved in AI development? 

During initial design. In our research, 87% of Advanced organizations bring cybersecurity, legal, and governance into the work during initial design, while 56% of Emerging organizations bring them in just before deployment or only when something comes up. When governance shows up at the end, the final review can send the team back to rebuild what it just finished. 

What is a player-coach technology leader? 

A player-coach sets direction and develops people while also weighing in on architecture, product decisions, governance, and delivery. They can review AI-generated output and tell whether it is good, they know who on the team is using AI well, and they understand what the tools cost. In our research, 42% of Advanced organizations say technology leaders spend more than half their time directly involved in execution, compared with 18% of Emerging organizations. 

Should companies remove layers of management when adopting AI? 

Be careful with flattening. AI can reduce layers of management, and removing a layer does not remove the coordination it handled. Unless that work has been automated, simplified, or clearly reassigned, it lands on whoever is left. Take the work out first, then move the people. 

Kortney K. Rupp is a Director at Paragon by Riviera Partners. Connect on LinkedIn.