Daniel Dines✓

When I started working on The Work That Remains, I wanted to challenge the narrative that AI models will eventually become so advanced that they will take over most of the work people do. Exploring this led me to a framework built around the strengths and limitations that both humans and AI have: AI proposes, humans decide, automation executes. For the new season of The Path Forward, I was joined by Cassie Kozyrkov @decisionleader to discuss how we build technology that enhances our ability to take initiative, exercise judgment, and make decisions. In a world where reasoning is becoming increasingly accessible, it is important to distinguish between intelligence and will. The assumption that models can simply optimize toward a given goal overlooks how complex processes actually work. Within an organization, countless smaller decisions and initiatives shape the outcome of a process. We discussed why the ability to take initiative is one of our most durable human traits and why it will become increasingly valuable. If we look at the decision-making process at a high level, we start with information, apply logical reasoning, reach a conclusion, and evaluate the consequences. AI excels in certain reasoning tasks, but it cannot bear responsibility for the outcome. When decisions carry significant consequences for an organization, people need to remain involved, exercising judgment and taking responsibility for what follows. Cassie made an important point about how we think of humans in the loop. Too often, it means placing someone in a position to simply approve or reject an AI recommendation, without necessarily enabling them to exercise meaningful judgment. I strongly agree. This is one of the central challenges of deployment: finding the right balance between using the reasoning capabilities of AI models and giving people the ability to evaluate, question, or change what AI proposes. Cassie, thank you for joining us. You can listen to our conversation here: YouTube: Spotify: Apple Podcasts:

The best AI model still doesn't know how your business actually runs. Here's how we're changing that.

The renewed debate about the pace of AI and related safety risks is extremely important, and I am glad it is happening between the labs that train the most capable models. UiPath does not train frontier models, so I will not put a number on the risk and I don't think anyone can. What UiPath does have is more than a decade of experience deploying automations in the most regulated environments there are: banks, insurers, hospitals, government agencies, and more. We have one job: making sure those automations do exactly what they are supposed to do, and nothing else. This experience is relevant to this debate. My advice? Slow down on what you cannot govern, and speed up on what you can. Most of your mission-critical business processes are in the second bucket. I wrote more on Substack ( and at length in The Work That Remains ( Let's keep building, carefully as always.

Years ago, it became clear to us that our customers' hardest problem would be coordinating work from beginning to end, and we started building for it. We bet on orchestration because that is where the challenge sits. A process has to exist in a form that agents and people can both act on, held and governed in one place, and coordinated with the systems doing the rest of the work. That layer is orchestration, and it is what we chose to build the @UiPath Platform around. Today, UiPath has been named a Leader in the @Gartner_inc Magic Quadrant™ for Business Orchestration and Automation Technologies. This one means a great deal to me. Execution is whether what we build holds up in production. Vision is whether we are building the right thing for what comes next. Being recognized in both says the direction was right, and that the market now sees it the same way. Thank you to the UiPath team for the years of work behind this, and to our customers for trusting us. You can read the full report here:

We rocked the BOAT. 🚣 Gartner® named us a Leader. Get the report:

My book is out today: The Work That Remains — Human Judgment, AI, and the Architecture of the Next Enterprise. I wrote it because most of what executives hear about AI comes from one of two rooms — one in panic, one in euphoria. Neither helps you decide anything. What enterprises need is a pragmatic account of where AI is genuinely strong, where it is structurally limited, and a deployment model that follows from both. That is what this book tries to be. My bias, declared up front: I run an automation company, and the conclusion is convenient for companies like mine. Do not take the argument on my authority — and do not dismiss it for the same reason. If you leave with one argument, take this one. AI genuinely embodies characteristics of human intelligence — it reasons, plans, drafts, explains. But it lacks the one thing every employee you ever hired does naturally: it does not learn on the job. A new hire arrives knowing nothing about your company and becomes valuable by absorbing everything nobody ever wrote down. The model arrives brilliant — and shows up exactly as new every morning. The entire method in this book is built around that limitation. I wrote it with Claude and ChatGPT, and the last chapter explains how. I learned more about AI writing this book than in any year of running an AI company. I think every reader will leave with something that changes how they deploy it. Read it here:

Every serious AI strategy eventually arrives at the same problem: the company was designed for people who could fill in the gaps. AI is forcing companies to expose the operating knowledge people have always supplied without being asked: what a customer promise actually commits the company to, which exception changes the path, who owns a decision, and what can safely happen next. Much of this never had to be formalized because people carried it through judgment, memory, and relationships. I keep hearing some version of the same hope: the next model release will automate the business. Better models will matter enormously, but intelligence is only one layer. The company still has to make the work legible by defining its objects, states, rules, permissions, gates, owners, and execution rails. That is what allows AI to contribute inside a system where people retain ownership of judgment and consequence. This changes where AI transformation should begin. Start with one meaningful workflow. Map the work, assign the rights, define the boundary between proposal, decision, and execution, and preserve the human knowledge the workflow has quietly depended on. Part IV of Between Panic and Euphoria, The Operating Model, is now live on my Substack. For leaders, it offers a sequence for designing the agentic enterprise. For employees, it offers a map of where durable work is moving.

You can read it here:

AI can absorb knowledge, reason within well-defined domains, and take on more of the work we once assumed only people could do. But knowledge is not identity. Memory is not experience. And capability is not motivation. In Episode 2 of The Path Forward, I sat down with @AndradaMorar to discuss the human side of the AI transformation and what people and companies need to preserve, develop, and rethink as the nature of work changes. Leaders need to look beyond the visible output of a role. Every organization has two ledgers. One records measurable outcomes: the contract reviewed, the case resolved, the deal closed. The other captures value that is harder to measure but just as important: the trust someone builds with a customer, the judgment developed through experience, the mentorship they provide, and the culture they carry. When companies treat AI only as a reason to compress jobs, they risk losing what does not appear in the first ledger. You can find Episode 2 on: Youtube: Spotify: Apple Podcasts:

In the first episode of The Path Forward, I sat down with Michael Atalla to talk about a question I keep hearing from customers: what comes after using AI as a productivity tool? Work does not usually begin with someone opening a chatbot. An invoice arrives. A customer request comes in. A case moves from one stage to another. A process has to continue, with the right context, controls, approvals, exceptions, and record of what happened. Michael made the point that no large enterprise runs on one stack. Work crosses clouds, applications, legacy systems, partners, and teams. That is why orchestration matters. It acts as the practical layer that connects people, agents, actions, workflows, systems, and governance into a business outcome. You can find episode 1 on: YouTube: Spotify: Apple Podcasts:

A friend of mine, a lawyer, told me she finally has the associate she always wanted. It drafts in seconds. It never gets tired. It never pushes back. Then she said the part that I always remember. The machine did not need her to have become anyone first. It arrived already good. She was not celebrating. The years of being wrong in front of a partner, of slowly becoming a lawyer, are the years the machine skips. And she wondered who, in her firm, will ever become her now. I have spent years building the layer that turns AI into work that gets done. I was convinced that intelligence plus action was the whole system. I was one layer too shallow. I am starting to write about what I think is actually missing, and why your AI strategy and your workforce are not two decisions but one. Part I is up. It is the start of a series I am calling Between Panic and Euphoria:

A smarter model does not automatically become a trusted participant in the business. It does not know by default where the boundaries are, what the consequences of an action might be, who is allowed to approve something, when to escalate, or what needs to be recorded for audit and security. The progress in AI has been extraordinary. Models reason better, agents write code, and systems generate and analyze increasingly complex work. So it is natural that leaders are now asking a more urgent question. How do we bring this intelligence closer to the real work of the enterprise? To make AI successful, the environment around it has to be engineered too. The business entities, the actions, the permissions, the controls, and the implications need to be clearly defined. That is why we started The Path Forward, a new @UiPath podcast. I see it as a place to think out loud with people who are close to this work, including our own teams, customers, partners, and external voices who are working through this shift in real organizations. I want the conversations to stay close to the work itself, exploring where AI helps, where it breaks, what people still need to decide, and what companies have to build before they can trust it in production.

Season 1 is now available on Spotify: Apple Podcasts: YouTube:

Today we announced our first quarter results, with ARR growing 12 percent year-over-year to $1.901 billion. One year into general availability, our agentic products are moving from pilots into real deployments, with customers standardizing on UiPath as the orchestration and automation execution layer for their enterprise AI transformation. The next step in that transformation is reducing the distance between an idea for automation and a governed system running it in production. With the launch of UiPath for Coding Agents, we bring coding agents into the full automation lifecycle. This changes the economics of implementation. It reduces the friction between what a customer wants to build and what they can deploy, govern, and scale. I’m grateful to the UiPath team for the focus and discipline behind this quarter, and to our customers who continue to use our platform to drive real transformation.

Our quarterly business update: 1Q ‘2027 revenue of $418M, up 17% year-over-year, with $1.901B in ARR, which is growing 12% year-over-year. Read more in our full earnings release:

As AI models become more capable, the question of what humans contribute uniquely becomes more pressing - and more interesting.

I felt this in Bengaluru at Devcon, surrounded by a community of builders that has shaped @UiPath from the beginning. Builders that take technology seriously enough to push it beyond the obvious use cases. While I was there, I had the great pleasure of welcoming @RomalShetty for a fireside chat on what this next era asks of people and organizations. One idea stayed with me: regardless of how much AI models evolve, there are certain human-skills that remain essential. Humans bring the initiative to decide what should be built, the judgment to know whether it works, and the ability to connect domain understanding, process knowledge, and technology. We also discussed why this makes the development of junior talent so important. Judgment is not formed in theory. It comes from mentorship, responsibility, and being close enough to the work to understand the consequences of decisions. The people who will guide AI-powered organizations tomorrow need the opportunity to build, question, and learn today. Romal, thank you for joining us. It was a great honor to have you at Devcon.

I’ve said it often because I believe it: coding agents will be as consequential for software as the cloud transition was. Despite the name, they don't just write code. They build. For the last few months, we've focused on a specific question: what happens after a coding agent finishes writing? How does that code become a working outcome inside an enterprise: secure, governed, running against real systems? Today we're announcing UiPath for Coding Agents. Builders can use the coding agent of their choice to create, test, and deploy automations across the full lifecycle on the UiPath platform.

More details on our announcement here:

If you started your company over today with two people. A seller and a builder. Who's the third hire?

That doesn't make AI less powerful, it just changes where the value sits. A year ago what we're building at @UiPath would have sounded like science fiction. Agents that interview subject matter experts, process documents, write the code, test it, deploy it, monitor it, handle exceptions. All of it. Today that's on our near-term roadmap. But even as we automate more, the human role gets more important. The judgment call that no model can make for you. Enterprises need to trust who they hand the keys to and that trust is built over years, not quarters.

Talked about all of this with @johnkoetsier and @matalla on TechFirst. Worth your time if you're thinking past the buzzwords.