The Path Forward: A leader’s guide to agentic transformation

I’ve spent my career working in B2B technology, starting with the advent of enterprise computing through the cloud era and the SaaS revolution, connecting software developers and buyers using brand-level storytelling built on innovative use cases and real results. While the outcome conversation is always the most important one, this AI moment feels different.   

With AI, this is the conversation I keep hearing about. Someone describes their AI initiative: multiple pilots running, solid vendor relationships, and enthusiastic teams. Everything looks promising. Then, the big question comes: “What’s actually changed for the business?”

Long pause.

Most enterprises have AI in their stack. Unlike those other transformational moments, the difference with AI isn’t the technology—which is exactly the point at which the conversation has fundamentally changed. The enterprises making real progress with AI aren’t better resourced or further along technically. They’ve just worked through the right things, in the right order. Those that are stuck haven’t.

That’s the premise behind “The Path Forward,” the new UiPath podcast series. The first five episodes, featuring candid conversations with UiPath leaders, customers, and industry thought leaders, lay out a practical map for enterprise AI transformation. Across these conversations with guests like UiPath Founder and CEO Daniel Dines, UiPath CMO Michael Atalla, Zack Kass, Author, Global AI Advisor, and Former Head of Go-to-Market at OpenAI, UiPath Chief Product and Technology Officer Raghu Malpani, UiPath leaders such as Simi Gupta and Andrada Morar, and enterprise customers like Omega Healthcare, you’ll hear what it takes to get from experimentation to impact, and what business orchestration and automation actually requires.

Here’s a peek at what you’ll hear:

Defining your AI strategy before someone else defines it for you

Most AI strategies aren’t strategies; they’re collections of use cases.

There’s nothing wrong with use cases; in fact, the building blocks of transformation often start with solving what’s right in front of you. But use cases alone don’t create transformation. Too often, they stay disconnected from one another, from the teams and systems that run the business, and from the desired outcomes leaders really care about.

In the first episode, Daniel lays out a four-stage framework describing the journey of AI and automation that most companies are on, and these stages can be the backbone for building a strategy:

• Stage 1: AI as a productivity tool. A human asks, AI answers.

• Stage 2: AI takes actions when a human initiates.

• Stage 3: an agent is initiated by an enterprise workflow, runs unattended, and produces real business outcomes.

• Stage 4: agents propose, execute, and decide when to escalate to a person.

Based on our experience working with enterprise customers, most companies think they’re building toward Stage 4 but are actually targeting Stage 2. And that’s important, because Stage 1 and 2 improve individual productivity, but very likely won’t change the trajectory of the business. Stage 3 is where many organizations begin to realize measurable enterprise impact… and you don’t get to Stage 3 without orchestration. 

When we talk about orchestration, we mean process orchestration. And that means orchestrating all the actors involved to deliver an outcome for the business—the people, the workflow itself, the automations, the actions, the business systems.

— Daniel Dines, Founder and CEO, UiPath

Orchestration is the coordination layer—the connective thread that makes AI agents, people, robots, and systems actually work together. An agent that can’t take an action in a real business system isn’t an agent. It’s a chatbot wearing a different hat.

One more thing the strategy has to get right from the start: openness. No enterprise runs on a single stack. 

As UiPath Chief Marketing Officer Michael Atalla put it, enterprise workflows “cross all of those boundaries”: multiple clouds, legacy and modern systems, plus line-of-business applications that weren’t built to work together. 

Any AI strategy that doesn’t understand these stages and account for the reality of enterprise workflows isn’t a strategy. It’s a plan to rebuild later.

The people question that most AI leaders are getting wrong

There’s a binary way of looking at the outcomes of AI adoption and what it means for the workforce. Either it’s going to unlock human potential at scale, or it’s going to hollow everything out. Neither framing is useful for the decisions in front of leaders right now.

Here’s how Dines frames it: there are two ledgers, and most organizations are only tracking one.

The visible ledger is straightforward: tasks automated, costs reduced, savings realized. One plus one plus one equals three, every time. But that doesn’t really account for the human potential aspect. Those pieces are found in the “hidden” ledger, capturing institutional memory, how things get done in the real world, and the people who make things work that aren’t in any process map. The ones who’ve been around long enough to know which rules to follow and which ones to push back on.

It’s that second ledger–equally as important but largely unseen and harder to quantify–that includes institutional memory, human judgement, customer context, company culture, and the informal map of how work actually gets done in the enterprise. It lives with the people who know which rules to follow, which ones to question, and how to navigate the exceptions that never show up in a process map. 

Leaders who only track the visible ledger risk what Dines calls a “hollow enterprise”: AI in every seat, but no one left who knows why decisions were made, what the company stands for, or how to read a room. That’s not transformation. That’s attrition with better branding.

In our second episode, Andrada Morar, VP of Customer Experience, Global Partnerships, Tech Alliances at UiPath, draws the line between what AI replaces and what it genuinely cannot:

Models have memories, but they lack the motivation to be excellent. AI can provide a lot of the knowledge that was seen as the skills before. But it will not give you the curiosity. It will not give you the grit to push through when something doesn’t work.

— Andrada Morar, VP of Customer Experience, Global Partnerships, Tech Alliances, UiPath

This matters for the future workforce, and the talented candidates that will make up that cohort. If organizations stop hiring early-career talent because they think AI can handle entry-level work, they risk cutting off that next wave of employees (and leaders) that will fill that second ledger, severing the pipeline that sustains culture, judgement, and growth over time. The stakes here are high, and choosing the right transformation path can either strengthen your organization or hollow it out.

Build for production, not proof of concept

Here’s the question nobody asks in a pilot: “Can this actually ship?”

Pilots always look successful when they’re contained. The failure to scale occurs when you move out of that demonstration environment and into a production one. Suddenly, the reality of your architecture and operational infrastructure get tested, and two patterns show up almost every time.

Pattern one: Every process and every step becomes a candidate for an AI agent. Everything becomes agentic, everything becomes cognitive, and the system becomes expensive, unpredictable, and difficult to audit at scale.

Pattern two: The agent capability gets built first, and then you try to wrap governance around a system that wasn’t designed for it. 

Malpani had an analogy that stuck with me:

Orchestration is the difference between a company owning a fleet of trucks versus owning a logistics company. The trucks do the work, but the logistics company decides what the trucks do, where they go, how they handle a flat tire or a driver falling sick or a flood happening in a route. Orchestration is the logistics company—and the workhorses are the tasks.

— Raghu Malpani, Chief Product and Technology Officer, UiPath

The principle he laid out is worth holding onto: not everything should be agentic. Where the process is deterministic, well-defined with a  predictable outcome? Keep it that way. Instead, apply cognitive agents where cognitive work actually happens. Not understanding what should be agentic and what should remain deterministic creates systems that are costly, unpredictable, and nearly impossible to scale.

What does this look like in a real business? Healthcare revenue cycle management is one of the most complex process environments that exists: thousands of payers, hundreds of treatment settings, insurance contracts negotiated employer by employer, with Medicare and Medicaid rules on top of all of it. The possible combinations are nearly infinite, and it doesn’t require AI everywhere. It does, however, require orchestration as a foundational piece, routing work across the organization, escalating where needed, and maintaining governance and auditability throughout. 

Adopt the leadership mindset this moment demands

There’s a version of AI transformation that fails for purely human reasons. The faultline is not technology or architecture, but because no one in the room can tell a believable story about where that transformation is headed. 

Zack Kass, Global AI Advisor, and Former Head of Go-to-Market at OpenAI, and author of “The Next Renaissance: AI and the Expansion of Human Potential,” has thought harder about this than almost anyone. In our fourth episode, he shared it in a way that stayed with me: 

It is so hard to say, ‘We’re going to automate things to the hilt and it’s going to be great’ if you don’t actually believe that. So you have to at some point pick apart your own ideas such that you can tell stories about a future that people you are leading will want to charge to.

— Zack Kass, Author, Global AI Advisor, and Former Head of Go-to-Market at OpenAI

That is the work of leadership now: making the future feel credible. People do not need another transformation slogan. They need transparency to understand what will change, what will stay anchored, where human judgment still matters, and why the organization will be stronger on the other side.

The workforce conversation also has an emotional dimension that does not get nearly enough attention. In our conversation, Zack pointed to what workers facing automation often value most about their jobs. Not only wages or job security, but community.

Leaders who talk only about economic trade-offs may miss what their people are truly trying to protect. The best leaders connect AI transformation to a more compelling version of work, one where people spend less time trapped in repetitive tasks and more time applying judgment, creativity, context, and care. The most human parts of work.

This is strategic work. People will not charge toward a future they cannot see, do not trust, or do not believe their leaders believe in.

Stop shopping for models and start building the foundation

The enterprise AI conversation is dominated by model comparisons. Which LLM is best? Fastest? Cheapest?

It’s the wrong conversation.

Simi Gupta, Director of Client Success for Financial Services and Regulated Industries at UiPath, puts it directly in episode five:

Model used to be the hardest part. It isn’t anymore. These days, the model is just a procurement decision. You can go out and find a really amazing model this afternoon. What you cannot find is the answers to what it is allowed to do, on what data, within what limits, and who is accountable when it acts. That’s all orchestration.

— Simi Gupta, Director of Client Success for Financial Services and Regulated Industries, UiPath

In a single regulated workflow, intelligence is roughly 10% of the work. The remaining 90% is coordination, guardrails, and audit trails across every system the workflow touches.

Here’s the one thing worth internalizing from every conversation on this topic: confidence is engineered, not discovered. You don’t wait to feel comfortable with a model. You build the operational scaffolding—controls, traceability, human touch points, a clear definition of what the model can and can’t do—and that’s what makes accountable deployment possible.

The goal isn’t autonomous AI. The goal is accountable AI. Every decision should be reconstructable. In regulated environments, regulators will tell you it didn’t happen if you can’t reconstruct it. But that accountability standard matters far beyond finance and healthcare. Trust in AI gets built or broken one decision at a time, and it can only be built if the decisions are traceable.

Getting this right requires clarity about who owns what. Simi’s model is simple:

“An agent without a business owner is a science project. An agent without IT is unsupportable. Without risk, it is unshippable.”

The companies that scale don’t get there by moving fast and hoping governance catches up. They design the controls, audit trails, and parameters first, then build capability within those boundaries. 

We’ve seen many organizations get stuck in the sandbox and have to rebuild when they try it the other way. 

The path forward

The five conversations described above don’t exist in a vacuum. They are layers that are dependent on one another when constructing an approach to AI transformation. Get strategy wrong and architecture follows you off a cliff. Ignore the people dimension and you hollow out the organization you’re trying to transform. Build for pilots instead of production and you’ll be building and rebuilding forever. Skip governance and nothing will ship.

Making real progress won’t come from having unlimited budgets, the flashiest models, or the most pilots. Success looks more like asking better questions, making clear and consistent choices, and building (and reinforcing) the operational discipline and long-term vision to turn AI into business outcomes.

Check out “The Path Forward” as we go deeper on each part of this enterprise transformation journey, hearing from the people navigating it everyday. It’s worth the listen.

Listen to The Path Forward →

YouTube

Spotify

Apple Podcasts