Daniel Dines

The through-line

Dines keeps returning to a single argument: intelligence has to sit on top of reliable execution, not replace it. He frames robotic process automation and agentic AI as versions of the same project, because both "look at existing manual processes and we are trying to automate them by imitate the people exactly how the people will build those processes" [1]. Where RPA captured keystrokes and mouse clicks in a logical order and repeated them forever, agents handle the parts that could never be pinned down deterministically, and for him that makes agentic AI "a natural extension" of the work UiPath has always done rather than a break from it [1]. The biological analogy he uses is deliberate: life started with simple organisms "capable of very reliable, predictive actions, and then intelligence came on the top of these capabilities" [2].

The second half of the through-line is orchestration. Deploying one bot or one agent is trivial; deploying hundreds means coordinating "between agents robots and people," and because AI is "a probabilistic technology," humans usually have to stay in the loop [1]. He describes having pivoted the entire company into building a new orchestration engine, and claims UiPath was "among the first that preached orchestration as being essential to deploy agentic AI" at a time when it was a small software category [2]. What has shifted over time is the scale of the claim: orchestration has gone from an internal deployment mechanism inherited from the RPA era to the thing he says wins competitive bake-offs.

On why AI pilots fail

Confronted with the MIT finding that 95 percent of companies see no ROI from AI pilots, Dines does not dispute it. His diagnosis starts with the missing foundation: "many companies are stuck into building chatbots that have little impact on productivity," when the outcomes only arrive if you "deliver as much autonomously as possible, the entire process automation" [2]. He is blunt about advisory-only agents: "An AI agent that doesn't do action is not an AI agent," and a local chatbot that saves ten minutes a day "will help only my dog, because I will walk it 10 minutes more, but it's not going to meaningfully impact the P&L" [2].

The other failure mode is timidity. Companies "go after small use cases, they might have mild successes, but because of the risks involved, they don't have the guts to put them in production" [2]. His counterintuitive advice is to pick the hardest, most painful, most energy-consuming process, because even a five percent time saving there excites the organisation and rallies people behind the project, whereas a clean win on something trivial persuades nobody [2]. He points to a customer, CSL, whose plasma donor process was strategically critical enough that success there moved the whole organisation [2]. He also tells buyers to choose "a vendor that puts skin in the game," on the grounds that there are no blueprints for large-scale implementations yet and they are being built now, and says UiPath is willing to entertain outcome-based deals because the industry is in its early innings [2].

On the advantage of an existing automation foundation

Customers with a mature RPA program are, in his account, structurally ahead: they have already identified their processes and the steps that can be reliably automated, so the remaining work is precisely the set of steps agents should attack [2]. He describes clients doing exactly this, looking at processes where robots are already deployed and "inserting AI agents" into the gaps, managed inside the same platform [1]. For companies without that base, his advice is to build one, and to merge the AI center of excellence with the automation center of excellence, "because in a way, they belong together" [2].

On orchestration patterns and human oversight

He sees two schools of thought competing: a master agent that dynamically chooses paths through a process, or a workflow on top with agents handling small pieces [1]. His own view is that both patterns will survive and the choice is domain-specific. Customer service, where the path is genuinely open, suits a master agent; processes like procure to pay or order to cash already have "a lot of fixed path" and gain nothing from one [1]. This is consistent with his broader bet that workflow automation becomes more important in the agentic era rather than less, since "every SaaS application in the end, it's a database, data schema and some workflows," and a database, an ontology and a good workflow engine with agentic on top can achieve a great deal [2].

On being the Switzerland of platforms

Openness is the commercial argument he leans on hardest. UiPath partners with open source agentic frameworks including longchain, llama index and crewi, so customers can build wherever they like while UiPath delivers those agents "securely uh with great governance and pair it with actions" inside its own orchestration [1]. He extends the same logic to foundation models and data: partnerships with OpenAI and Gemini, with Snowflake's Cortex, and with NVIDIA around Nemotron, plus a stated intention to keep adding data providers through well-crafted connectors rather than becoming a data provider itself [2]. The point of all of it is to stop customers "feeling locked in into a particular platform" [2]. He says UiPath is winning bake-offs against Salesforce, ServiceNow and Microsoft, and when it wins, "it's because of our orchestration story and the integrated story, the Switzerland of different platforms story" [2]. The rest of the trust argument is longevity and reliability: relationships with partners and customers running close to ten years, a platform that "delivers tangible results," and more than 10,000 enterprise customers [1][2].

On bringing the workforce along

Dines treats adoption as a cultural problem before a technical one, and describes UiPath running itself as customer zero. Every part of the organisation had to bring AI ideas, and those ideas became part of the compensation structure: "We show both the carrot and the stick in doing this" [2]. Performance talk units tied to agentic success cascaded from his direct reports through the organisation [2]. The framing he chose matters as much as the incentives: "This is not an exercise about reducing the workforce, but it's more of an exercise of growing without increasing the workforce" [2]. His worked example is customer success, which traditionally can only serve the top fraction of accounts; an agent emulating most of a CSM's work lets you reach the long tail, and the same logic applies to BDRs [2]. He acknowledges the fear and inertia directly, and argues people come around once they see the work shifts toward "more human-like tasks," better exception handling and deeper customer engagement [2].

On horizontal platforms, vertical agents and where startups should play

His "biggest learning" from the RPA years is that horizontal platforms are extremely hard to build, and that a decisive ingredient in UiPath's success was arriving early enough to be associated with the category name itself [1]. He does not think that trick repeats: AI is far too big a space for any one player to own, so "the best way to get into AI is to to get into verticaliz agents, find something that you really deeply understand the business process and add value" and then attach to one of the few horizontal platforms that emerge [1]. He observes that many startups have a distribution problem rather than a technical one, and offers UiPath's installed base as the remedy, citing the acquisition of a UK vertical AI company, PIK, doing price and inventory management, which had "a GTM problem not not a technical problem" and succeeded once deployed across a much larger base [1]. His venture vehicle Crew Capital, co-founded with Brandon Deer, backs entrepreneurs building category-defining B2B technology companies [4].

On how the software market changes

He does not expect agentic AI to displace SaaS so much as squeeze it. With a handful of kids using Cursor or Lovable able to assemble a business application in days, pricing pressure on traditional SaaS vendors is inevitable, but he thinks people badly underestimate the difficulty of building something "compliant, secure, with the proper governance" that integrates into a large enterprise stack [2]. Hence his timeline: "I don't see a day one major shift, but I see five- year major shifts in the making right now" [2]. What agentic actually does is reduce human input in processes so fewer people serve more customers, improve integration between SaaS providers, and push toward consolidation [2]. On market size he is unequivocal that the agentic opportunity is huge and that UiPath has "a right to win" a portion of it. His stated ambition is to build an agentic business at least as large as the RPA one, which reached a billion dollars in five years, and he expects a similar acceleration curve, recalling ARR of 5 million at the end of 2016, then 45, 170, 360, and almost 600 in 2025 [2]. Internally he tracks agents created, agents live per day and per month, user engagement, pipeline creation and agentic units sold [2].

On Europe, government and regulation

Dines is openly proud of his European roots while critical of European habits. The trap for European entrepreneurs, he argues, is staying inside Europe and especially inside their own country: founders in markets large enough to be comfortable, like France, Germany or the UK, work closely with local customers and limit themselves [1]. UiPath's counterexample is that Romania "doesn't have its own market basically and we were global in from day one," and the first real scale came not from Europe or the US but from Japan, through partners including Sumitosui bank, where orchestration at scale was genuinely built [1]. On the public sector he is sharper still: governments such as Singapore and Dubai are far more active in automation, and UiPath sells much more to US federal and UK buyers than to the European Union or European Commission, which he calls "a pity" [1]. His prescription is to embrace it from the top rather than legislate it: "it's it's less needed you know to regulate it rather than to adopt it accept it," otherwise Europe gets left behind [1]. Positioned between China and the US, Europe has to build its own destiny, and he singles out defense as the industry where European companies should be connecting themselves [1].

On speed as the only playbook

Asked what he would tell someone starting now, his answer is austere: "you have to run as fast as you can. It's the fastest company will always win. Just pick up your space. Go fast" [1]. Failure carries no penalty in his view, even in Europe, where he sees failure starting to be embraced, and it certainly does not end a career [1]. The danger is internal: "The moment when you start to doubt yourself are the moments of stagnation," because if you are not moving faster, someone else is [1]. For UiPath itself, sitting on incumbency, deep process knowledge and long customer relationships, the focus he names is simply execution [1].

Takeaways

• Build the automation foundation before adding intelligence; chatbots that save ten minutes a day do not move the P&L, and "An AI agent that doesn't do action is not an AI agent" [2].
• Aim high on purpose: attack the most painful process rather than accumulate small wins, because a five percent saving on something excruciating rallies an organisation where a clean win on something trivial does not [2].
• Merge the AI center of excellence with the automation center of excellence; they belong together, and existing RPA programs have already mapped the processes agents should extend [2].
• Expect both orchestration patterns to persist: master agents where paths are open, such as customer service, and workflow-first designs in fixed-path processes like procure to pay and order to cash [1].
• Do not try to build a horizontal AI platform; go vertical with deep process knowledge and attach to one of the few horizontal platforms, since most startups have a distribution problem rather than a technical one [1].
• Frame internal AI adoption as growing reach without growing headcount, and back it with compensation tied to AI ideas and agentic results [2].
• Agentic AI will pressure SaaS pricing rather than displace SaaS quickly; the real shift plays out over about five years because compliant, secure, enterprise-integrated software remains hard [2].
• Europe should adopt rather than regulate, sell globally from day one, and look to defense; Singapore, Dubai, US federal and UK buyers are already further ahead than the European Commission [1].