Azure AI. Tableau AI. UiPath Autopilot. Three launches, three eras, three completely different markets — and one lesson that tied them all together.
Three AI product launches across Microsoft, Salesforce, and UiPath reveal a consistent pattern: the technology is never the real problem. At Microsoft (2017-2019), the battle was making AI legible to non-technical buyers. At Salesforce/Tableau (2021-2023), the battle was overcoming the "demo vs. reality" skepticism. At UiPath (2023-present), the battle is trust and governance — buyers want to know what happens when the agent fails, not just what happens when it works. The through-line across all three: enterprise AI adoption is always slower than the market expects and faster than the skeptics predict, and positioning that acknowledges the messiness of implementation always outperforms positioning that only shows the highlight reel.
In 2017, I was writing positioning documents for Azure Machine Learning. The word "AI" was just beginning to feel safe to say in an enterprise sales conversation without someone rolling their eyes. We were careful. We led with "intelligent cloud." We let the demos do the heavy lifting. We were terrified of overpromising.
In 2022, I was relaunching Tableau as an AI-first analytics platform. The word "AI" was now mandatory. If it wasn't in your category description, investors and analysts assumed you were behind. We led with AI. We put it in the product name. We still let the demos do the heavy lifting. We were still terrified of overpromising — but for different reasons.
In 2024, I'm positioning UiPath Autopilot as the agentic automation layer for the enterprise. The word "AI" is now a liability in some rooms. Too generic. Too commoditized. Too associated with things that didn't work. We lead with "agentic." We lead with outcomes. We lead with governance. And the demo — the one that shows what happens when the agent fails gracefully and escalates to a human — is more important than the one that shows what happens when everything goes perfectly.
Same job. Three different companies. Three completely different conversations. Here's what I learned from each one.
The core challenge at Microsoft was not the technology. Azure's AI capabilities in 2017-2018 were genuinely ahead of where most enterprises were ready to use them. The challenge was translation. We were trying to sell machine learning to buyers who didn't have data science teams, couldn't explain neural networks to their CFOs, and were nervous about putting their most sensitive business data into a cloud model they didn't fully understand.
The PMM breakthrough — and it took us longer than it should have — was to stop selling the AI and start selling the outcome. Not "Azure Machine Learning enables predictive modeling at scale" but "your procurement team can predict supplier delays before they happen." Not "cognitive services with computer vision APIs" but "your quality control process can catch defects your inspectors miss."
The metric that changed everything was the Azure Synapse relaunch, where we drove a 400% year-over-year increase in digital impressions and 15% month-over-month revenue growth. Not by explaining the technology better. By finding the decision-maker — not the data engineer, not the IT architect, but the business unit leader who owned the process — and speaking in the language of what they were actually trying to fix.
The lesson from this era: enterprise AI adoption is always gated by organizational readiness, not technological readiness. The technology was ready before the org charts, the workflows, and the governance structures were. Positioning that acknowledged this — that sold "the on-ramp" as much as "the destination" — worked. Positioning that assumed buyers were ready to transform immediately almost always underperformed.
By 2021, the enterprise AI landscape had shifted fundamentally. Buyers had been to dozens of AI demos. They had been wowed in conference rooms and then watched implementations stall or fail. A new form of skepticism had emerged — not "is this AI real?" but "will it actually work in our environment, with our messy data, in the hands of our actual users?"
Relaunching Tableau as AI-first was harder than launching a new product would have been, for one reason: Tableau had a customer base that loved it for what it already was. Our loyal users had workflows, certifications, muscle memory. Telling them the product was now fundamentally different risked alienating the people who were already evangelical about it.
The positioning we landed on — and this took multiple iterations — was not "Tableau is now an AI product." It was "AI makes Tableau do more of what Tableau was always trying to do." Every analyst in your organization can have the insights that only your best analysts could find before. The AI wasn't replacing the Tableau experience. It was democratizing it.
That reframe changed everything. Pipeline exceeded targets by 150%. Close rates hit records. The field, which had been nervous about how to sell an AI-first story to existing customers, suddenly had a narrative they could walk into any renewal conversation with.
The lesson from this era: when you're relaunching something people already know, you're not selling a new product — you're selling an evolution of a relationship. The AI is the mechanism. The value proposition has to be about the customer's existing ambition, newly achievable. Positioning that centered on what the AI could do was less effective than positioning centered on what the customer could now do that they couldn't before.
The conversation I have most often in 2024 and 2025 with enterprise CIOs is not "does this AI work?" It's "what happens when it doesn't?"
This is new. This is the product of two years of enterprises deploying AI agents in production and learning the hard way that failure modes matter as much as success modes. An AI agent that works 95% of the time and fails silently the other 5% is not a reliable business process. An AI agent that works 95% of the time and escalates clearly, logs everything, and gives a human the context they need to complete the remaining 5% — that's something you can actually run a business on.
Positioning UiPath Autopilot has required building an entirely new vocabulary for enterprise buyers. The word "autonomous" — which every AI company wanted to use in 2023 — turns out to be the wrong word for most enterprise buyers. It implies things running without oversight in a world where oversight is what their governance frameworks, their compliance teams, and their boards are asking for.
The word we've landed on, and that I believe is right both strategically and technically, is "agentic." Agents act. They perceive their environment, make decisions, and take steps. But agents operate within a system — with guardrails, with escalation paths, with audit trails. "Agentic automation" tells a buyer: you get the intelligence and the action, without surrendering the control.
The demo that closes deals in 2025 is not the one where the agent completes a complex process perfectly from start to finish. It's the one where the agent hits an ambiguous situation, makes the right call about when it's outside its confidence threshold, surfaces the issue to a human with full context, and the human resolves it in thirty seconds. That's the demo that makes a CIO say "I could actually deploy this."
"Enterprise buyers aren't asking whether your AI is intelligent. They're asking whether it's trustworthy. Those are different questions, and they require different answers."
Here it is, the through-line I promised at the top: enterprise AI adoption is always slower than the market narrative and faster than the skeptics predict. And the PMMs who do best are the ones who build their positioning around the actual adoption timeline, not the hype cycle timeline.
In 2017, we could have positioned Azure AI as "the future of enterprise computing, available now." We would have gotten great press coverage and terrible pipeline. Instead we positioned it as the intelligent layer you add to the workflows you already trust, starting with the one business problem you most need to fix. That worked.
In 2022, we could have positioned Tableau AI as "the death of traditional BI." We would have spooked our installed base and given our competitors an easy attack vector. Instead we positioned it as Tableau getting better at the thing Tableau was always for. That worked.
In 2024, we could position UiPath Autopilot as fully autonomous business process execution. We'd get the AI-maximalist audience and lose everyone who actually needs to deploy this in a regulated industry. Instead we position it as the enterprise execution layer that makes AI useful in production — not in a demo. That's working.
The pattern is the same every time: meet the buyer where they are, not where you wish they were. Acknowledge the real fear — not to validate it, but to address it directly. And remember that the most important sentence in any enterprise AI pitch is not the one about what the product can do. It's the one that says: here's exactly how this fits into what you're already doing, here's what changes, and here's what doesn't.
That sentence is the hardest one to write. It's always been the hardest one to write. It doesn't get easier just because the technology gets better.
Kuber Sharma leads platform product marketing at UiPath, where he is responsible for the market narrative around agentic automation. Previously he led PMM for Tableau AI and Salesforce Data Cloud at Salesforce, and for Azure AI, Azure Synapse, and Azure Machine Learning at Microsoft. He writes Positioned, a newsletter on AI-era product marketing strategy.