revAlation 2026 Keynote: AI Velocity & Data Governance

Data and AI leaders from North America recently came to Chicago for revAlation. Behind them sit hundreds of organizations across almost every industry in the world, and a community of more than 18,000 people online. Alation CEO Satyen Sangani opened by pointing past the room: the event, he said, was never really about the people in it, but about the people who weren't — the ones everyone present works for and works with to build their organizations.

He described arriving each year with the same problem to solve: what the market needs from Alation, where it is going, and where the company chooses not to go. Strategy, in his framing, is ultimately a function of choice, which is why he began recalling the choices he made a year ago.

How AI has transformed data teams in the past year

Last year's prediction was that data teams would become agent builders, which was partially right.

Agents looked easy to build: write a prompt, associate it with a model, connect it with some tools, and you have an agent that does things. Even then, the easy part was declaring intent. The hard part was ensuring the agent did the right things, which hinges on two core capabilities Alation spent a decade building:

• Governed, accurate data: If the underlying data is flawed, the output is wrong, no matter how well the model was trained.

Governed, accurate data: If the underlying data is flawed, the output is wrong, no matter how well the model was trained.

• Context: The rules of the road that guide how the agent operates. If those two things drive accuracy, the teams who own them can become the builders.

Context: The rules of the road that guide how the agent operates. If those two things drive accuracy, the teams who own them can become the builders.

Two questions to the room tested the premise. Sangani asked how many had used chat to ask a question of their data, and every hand went up. When he asked how many had built a dashboard with a couple of prompts in Claude or ChatGPT, about half the room raised their hands.

Adoption, in other words, has happened, and it’s happening in the analytical work these teams already do. What Sangani said he underestimated was trust. Automation is moving more slowly than he predicted for a simple reason: nobody trusts AI completely. Sixty or seventy percent, perhaps, but not enough to let it finish the job.

The wider evidence points the same way: only 5% of organizations let agents execute high-stakes decisions without human review, and 60% cap them at moderate-risk task automation.

That caution, Sangani reasoned, is rational. When AI does the job, there is a whole set of subtle things it fails to capture, and those subtleties push the output slightly off. That slight offness means it can't be trusted to interpret the world exactly the way its user would, so the user keeps doing the work.

Why more AI output hasn't made teams better aligned

Sangani returned to a cartoon he keeps coming back to. One person says they took a bullet point and turned it into a long email with AI. The person receiving it says they took the long email and turned it back into a single bullet point:

It describes his own working life, he said. Two to three hundred emails arrive every day, against maybe 150 relevant ones before AI, and much of that content is now machine-generated. Documents to read and review went from two or three a day to roughly ten, all beautifully formatted, all with perfect outlines. The question he kept asking was whether wading through that left him in a better position than the earlier material, which was at least prepared by hand.

The same thing is happening with dashboards. The kind of dashboard that used to be a specialist craft can now come out of a general-purpose model in thirty minutes and a couple of prompts. Sometimes two.

None of it has made the fundamental work easier, he argued. Everyone still has to know what things mean inside their own business. When colleagues talk to each other, there is no more ease or simplicity in trying to align, because there are now not ten different definitions but a hundred. Everybody can be a builder. Everybody can be a creator. What none of that supplied was a better way to align, or to learn from each other.

Which is why, he said, he has become careful about semantic layers. Definitions, rules, and guardrails are all real, and all of them are downstream of something else. Alignment is about getting people in a room and making sure that what one person means is what the other means, and that what the person three doors down says is something everyone understands.

"Cheap production has multiplied what we have to reconcile, without giving us any new way to reconcile it."- Satyen Sangani

"Cheap production has multiplied what we have to reconcile, without giving us any new way to reconcile it."

- Satyen Sangani

If reconciling all of that is the real job now, then the deeper, harder job is learning: deciding what an organization believes, writing it down, and keeping it current as the business moves. That doesn't happen because everyone got faster individually. It happens when an organization learns deliberately, which turns out to be a different thing entirely.

Individual learning versus organizational learning

Individual learning, on his account, is going well. People are using these tools, being creative, working forward. The 10x learners and the 10x doers are visible, and what they can achieve is remarkable.

Organizational learning has a different character. It means the company holds a common set of facts, so that two people asking the same question get the same answer, or close enough that the difference doesn't change what either of them does next. That is a much higher bar than it sounds. The question is not whether the best people are getting faster; they are. It is whether the gap between the fastest and the slowest is closing or widening. Right now, Sangani argued, cheap AI output is widening it, because every new artifact is one more thing the organization has to reconcile. Organizational learning is about pulling everybody along at a consistent pace.

McKinsey's 2026 State of AI survey found that nearly nine in ten respondents report regular AI use in at least one business function, while 44% say AI is scaling across the enterprise, up from 38% a year earlier. McKinsey's own read is that broadening AI use may require redesigning workflows around AI and building platforms that can operate at scale.

Sangani would put it slightly differently, or at least add to it. What he sees blocking companies is data and context, data and knowledge. Gathering those two things together is the real impediment to making AI work. Even the organizations doing genuinely remarkable work, he noted, will say they are at the very beginning of their journey.

Why the data catalog should disappear

Anyone who watched the keynote also saw Alation’s recent rebrand and repositioning. The bowtie is gone, and we’ve leaned into black as our predominant color on stage, with traces of orange.

Sangani presented that as a design decision with an argument behind it. The old Alation was a destination, modeled on Wikipedia and Stack Overflow, where people produced and consumed knowledge in one place. That model assumed you would go somewhere to look something up. Fewer people do that now. The Wikimedia Foundation reported human pageviews to Wikipedia down roughly 8% year over year, and attributed the decline to generative AI and social platforms answering questions directly rather than sending people to the source.

Great design disappears, he said. Which means Alation needs to effectively become an invisible substrate, showing up inside the code editors and assistants where work already happens.

The company learned why that matters by getting it wrong internally. We gave our own teams raw AI tools and got sprawl: duplicate definitions, unmanaged artifacts, and real token costs. At one point we discovered different internal dashboards tracking the same metrics that didn't fully align.

The lesson was not that dashboards are the problem. Dashboards aren't going away, because people need to measure performance and learn. Static dashboards are the problem. So we built Intelligent Feeds: governed, interactive intelligence that watches the underlying sources and suggests how the artifact should change. Our sales messaging now evolves off support logs, customer feedback, and competitive movement, rather than sitting in a deck until someone remembers to revise it.

Three properties, he said, make that safe to build on:

• Governed, so it can be trusted.

Governed, so it can be trusted.

• Open, so it works with whatever stack you already have.

Open, so it works with whatever stack you already have.

• Self-improving, so corrections compound.

Self-improving, so corrections compound.

The last one is the one people underrate, and Sangani credited a sharper view of why to an Imagination In Action panel at Google Bay View. Liran Zvibel of WEKA described what breaks in nested architectures, where agents call other agents — turtles all the way down. A nested agent operates on a completely different context than the one that called it, so every step holds up on its own, and the chain still produces something a human would look at and reject. Melissa Valentine of Stanford named what has to sit at the bottom of all those loops: a layer of verified, typically human-created data for the evals. She was also honest about why companies skip it: Defining success precisely enough that humans can apply it consistently is slow, unglamorous work, and most people would rather not do it.

This is the mechanism by which we ultimately push AI to trusted action: Loops of agents with human judgment underneath holding them up. If you can't trust the ground truth at the bottom, none of the loops above it mean anything.We wrote up that panel in more detail here.

Data governance and BI teams are about to converge

Sangani's forecast for 2027 is that governance teams will increasingly merge with BI teams. When building a report is a matter of writing and refining English, the person who produces the information and the person who vouches for it stop being different people. That collapsing boundary between creation and validation arrives at a pointed moment: 78% of senior leaders lack strong confidence they could pass an independent AI governance audit within 90 days.

He may be wrong about the timing again, he said. He doesn't think he is wrong about the direction.

Where to begin

The question Alation gets most often is where to start, and the answer was one foot in front of the other. Three customers took the stage to walk the audience through very different first steps:

• Electronic Arts, a global leader in digital interactive entertainment and a massive video game publisher and developer moved quickly from catalog to data products at scale, built agents on top of them, and used ontologies to lift model evaluation scores.

Electronic Arts, a global leader in digital interactive entertainment and a massive video game publisher and developer moved quickly from catalog to data products at scale, built agents on top of them, and used ontologies to lift model evaluation scores.

• Water Technologies activities of Veolia, a multinational specialist in water and wastewater treatment solutions for industrial clients and public authorities took the structured route: catalog, then data quality, then layered abstractions over the estate.

Water Technologies activities of Veolia, a multinational specialist in water and wastewater treatment solutions for industrial clients and public authorities took the structured route: catalog, then data quality, then layered abstractions over the estate.

• Ally Financial Services, a leading digital financial services company and bank holding corporation, consolidated its most critical data first and scaled knowledge from there.

Ally Financial Services, a leading digital financial services company and bank holding corporation, consolidated its most critical data first and scaled knowledge from there.

None of them tried to do everything. All of them picked something that mattered and made it true.

This moment belongs to data teams. Whether the work is fighting threat actors or curating a definition, it comes down to one fundamental question: what is true, and what do we know? An organization that cannot align around facts becomes paralyzed.

Pick a problem, pick an intelligence feed, and let's build it together. Contact us today to get started.

Sources & notes

Every external claim in this post is independently verifiable. The public sources are listed here.

• Only 5% of organizations permit agents to execute high-stakes decisions without human review; 60% limit agents to moderate-risk task automation. Survey of 950 C-suite and senior business leaders across 10 industries in the United States, fielded 23 February – 18 March 2026. — Grant Thornton, 2026 AI Impact Survey — https://www.grantthornton.com/services/advisory-services/artificial-intelligence/2026-ai-impact-survey

Only 5% of organizations permit agents to execute high-stakes decisions without human review; 60% limit agents to moderate-risk task automation. Survey of 950 C-suite and senior business leaders across 10 industries in the United States, fielded 23 February – 18 March 2026. — Grant Thornton, 2026 AI Impact Survey —

• Nearly nine in ten respondents report regular AI use in at least one business function; 44% report AI scaling across the enterprise, up from 38% a year earlier. Survey of 1,719 participants in 97 nations, fielded 4 May – 8 June 2026. — McKinsey, The state of AI in 2026: On the road to ROI, 25 August 2026 — https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

Nearly nine in ten respondents report regular AI use in at least one business function; 44% report AI scaling across the enterprise, up from 38% a year earlier. Survey of 1,719 participants in 97 nations, fielded 4 May – 8 June 2026. — McKinsey, The state of AI in 2026: On the road to ROI, 25 August 2026 —

• Broadening AI use may require businesses to redesign workflows around AI capabilities and to establish platforms that allow them to operate at scale. — McKinsey, AI at work but not at scale, 10 December 2025 — https://www.mckinsey.com/featured-insights/charts/ai-at-work-but-not-at-scale

Broadening AI use may require businesses to redesign workflows around AI capabilities and to establish platforms that allow them to operate at scale. — McKinsey, AI at work but not at scale, 10 December 2025 —

• Human pageviews to Wikipedia declined roughly 8% against the same months in 2024, which the Foundation attributes to the impact of generative AI and social media on how people seek information, particularly search engines answering questions directly. The Foundation notes that because it revised its bot-detection logic mid-2025 and reclassified traffic for March–August 2025, the revised data should be interpreted with care. — Wikimedia Foundation, New User Trends on Wikipedia, 17 October 2025 — https://diff.wikimedia.org/2025/10/17/new-user-trends-on-wikipedia/

Human pageviews to Wikipedia declined roughly 8% against the same months in 2024, which the Foundation attributes to the impact of generative AI and social media on how people seek information, particularly search engines answering questions directly. The Foundation notes that because it revised its bot-detection logic mid-2025 and reclassified traffic for March–August 2025, the revised data should be interpreted with care. — Wikimedia Foundation, New User Trends on Wikipedia, 17 October 2025 —

• Panel "Engineering the Agentic Enterprise," Imagination In Action, Google Bay View, Mountain View, 14 September 2026 — https://imaginationinaction.co/2609sv/14. revAlation Chicago was held 16–17 September 2026 at The Old Post Office, Chicago —https://www.alation.com/revalation/chicago/

Panel "Engineering the Agentic Enterprise," Imagination In Action, Google Bay View, Mountain View, 14 September 2026 — . revAlation Chicago was held 16–17 September 2026 at The Old Post Office, Chicago —https://www.alation.com/revalation/chicago/

• Liran Zvibel, Co-Founder and Chief Executive Officer, WEKA — https://www.weka.io/company/leadership

Liran Zvibel, Co-Founder and Chief Executive Officer, WEKA —

• Melissa Valentine, Associate Professor of Management Science and Engineering, Stanford University, and Senior Fellow, Stanford Institute for Human-Centered Artificial Intelligence — https://profiles.stanford.edu/melissa-valentine

Melissa Valentine, Associate Professor of Management Science and Engineering, Stanford University, and Senior Fellow, Stanford Institute for Human-Centered Artificial Intelligence —

• 78% of business executives lack strong confidence they could pass an independent AI governance audit within 90 days. Survey of 950 C-suite and senior business leaders across 10 industries in the United States, fielded 23 February – 18 March 2026. — Grant Thornton, 2026 AI Impact Survey —

78% of business executives lack strong confidence they could pass an independent AI governance audit within 90 days. Survey of 950 C-suite and senior business leaders across 10 industries in the United States, fielded 23 February – 18 March 2026. — Grant Thornton, 2026 AI Impact Survey —