The State of Analytics, as Told by the 2026 Tableau Conference

Welcome back to yet another installment of “Ann’s takeaways and observations” from Tableau’s annual conference. This marks the fourth year that I’ll be sharing with you what I witnessed at their user conference in long form. You can find historical stories here: 2023, 2024, and 2025.

Let me start plainly: I did not attend the conference in person this year. I was given the opportunity to attend and participate, but I declined.

I did however “attend” from a distance, that is, watched their keynote presentation, watched Devs on Stage, and observed the DataFam react on their social media platforms. Playing the part of lowkey lurker that I’m not publicly known for.

So below you’ll find my perspective on the direction Tableau is going, the direction analytics is going, my reaction, and a familiar technical appendix at the end filled with an assessment of what was shown at Devs on Stage.

Sitting at my desk at home watching the keynote is definitely different than the prior 9 years, not least of all because this year there’s no longer a CEO of Tableau. Instead we’ve got Southard Jones, the CPO at the top of the bill, shouting “Good Morning DataFam!”

We spend five minutes asking different groups of people to stand up. And I can only imagine the collective relief when he finds that asking “first timers” to stand up reveals that 40% of the audience hasn’t been there before.

They’re new to this, so the act of shouting “DataFam” like you know them can hold. Maybe they don’t know that he’s only ever known TC as it appears in San Diego. Maybe for them it lands as typical gimmicky marketing jargon and not as a label bestowed upon a group of people whose ears it actively grates against.

And then he launches into his speech, which reveals something that slides off the tongue effortlessly, but reveals so much. That everyone in attendance is here to do what they want to do, “to help people see, understand, and act on data.” The reframe of the mission statement crafted back in Seattle, where Salesforce definitely wasn’t in the room.

Now this addition of “act on data” isn’t new, it is something they’ve been workshopping for 4+ years, but seeing it now, restated in the keynote, reminds me of the state of play. We’re here to discuss how we can “Unlock the Power of Agentic Data,” presumably using AI agents to take action. So much for the humans.

No, I take that back, because on the very next turn, he insists “there’s one thing AI cannot do… you know why people trust the dashboard… because someone here has gone to great lengths to make sure it is correct.” So we’ve got the human back, as the gatekeeper of data, controlling the data, ensuring the joins are correct, so that AI agents can then use the data.

As that devastating truth lands, the practitioner’s new role is revealed. Not as the central sensemaker, but as the cog whose job it is to prepare data for the machines to make use of. It is also at this point that I remember the quintessential style that Salesforce likes to maintain in these keynotes, the concept of eras and chapters.

Before we go too deep though, we need to meet the brand-new, 55-days-into-the-role General Manager, a fact he insists on repeating, Mark Recher. The introduction of a GM serves as another signal to the audience. Tableau no longer needs a CEO, it merely needs a GM, because its future now clearly resides as an operational layer inside of Salesforce. Not a standalone product, not a standalone platform, the data-oriented guts of Salesforce.

As he touts the statistic that “89% of leaders say they’ve experienced inaccurate or misleading AI outputs,” he calls attention to the fact that they’ve now got a clearer picture on the agentic analytics era. Last year we called you an “Outcome Whisperer” this year, you’re a “Knowledge Architect.” Last year we called you an “AI Advisor,” this year you’re an “Agentic Architect.” Last year you were a “Trust Enabler,” this year a “Decisions Architect.”

And while I’d much prefer to put architect on my LinkedIn profile over whisperer, we’ve got our framing of the actual chapters to come, revealed moments later: architecting knowledge, powering decisions, and agentifying actions. Here is the heart of what this keynote is aiming to convey to the audience. The future of Tableau, chapter by chapter.

Before sitting down, Mark makes his boldest statement, through this technology you now “have the power to tell your organization what they need to know before they even understand it.” A statement that lands like an assassin, silently invalidating the original mission statement. An assassin whose cloaked actions claim another innocent, the 2015 data analyst inside me who just discovered Tableau and its mission statement for the first time.

To help workshop the chapters, Salesforce’s own Chief Data Officer, Michael Andrew, is brought out. He shares that it took 40 emails plus one directly to the CIO to get Tableau installed at Salesforce. A charming story wrapped up by the convenient acquisition of Tableau mere months later. A story that buries the most important metric of the show: it took 41 emails to install the very technology that you, at your own organization, are supposed to be advocating for.

Beyond the institutional lore, the CDO shows off composable data sources, a soon-to-be-available knowledge graph in Cloud and Server, and various methods of how AI is infused throughout all of Tableau, especially Tableau Next. These are the pillars by which you will architect knowledge.

As the CDO departs, we launch into a snappy video of Engine, showing “how they use Tableau to succeed.” Engine seems to be a modern alternative for corporate travel booking and management. A genius move for the company, who finds itself in a primed audience who has just experienced the pain of corporate travel to be at the event.

What we don’t see in the video is anything resembling Tableau Classic, instead we see that Salesforce and later Slack are heralded as the operating system for the company. Tableau’s there by way of agentic analytics agents, natural language joins, metrics inside of Slack, and bar charts in chatbots.

As our attention is drawn back to the stage, Tableau’s CMO Rekha Srivatsan, appears to guide us through the next chapter. She begins with “You know what’s the catch? Unfortunately with access to all the AI tools, all the reports, all the dashboards, the whole decision-making process itself is slower, it’s harder, and it’s more disconnected than it needs to be.”

She’s not wrong there, but in saying this out loud, she is also silently saying: your modern BI stack of Snowflake + dbt + Tableau (with a dash of Collibra) is slower, harder, and more disconnected than it needs to be. A statement whose hypothesis only becomes true if you believe an entirely Salesforce-centric approach to all your institutional data is the fastest path forward to agentic analytics nirvana.

A hard pill to swallow for those corporate analytics teams that are just now feeling comfortable with their team structures and roles, those that have embraced the term analytics engineering while keeping visualization intact. An even harder pill for someone to swallow before analyzing even a drop of data.

Rekha is a gold mine, because in her next sentence she pulls out the most confident statistic I’ve ever heard. The premise: “if you ask the same question of different AI tools, 99.9% of the time they will give you back a different answer.” In reality she is describing the very means by which large language models both work and are differentiated.

LLMs are prediction engines with built-in, controlled randomness. They also contain varying bodies of knowledge and instruction sets by design. The phenomenon she’s describing is something you can experience by asking the same question of ChatGPT, Gemini, and Claude. Or by simply clicking the regenerate response button.

At this point I must apologize. Earlier when I said I declined being in attendance isn’t quite the full picture. Because I’m actually there, on Rekha’s slide as part of the DataFam AI showcase.

Yes, that’s right. In the fall of 2025 I won $9,000 and the prize of “most impactful” in a Tableau Next hackathon. My award-winning submission, a philosophical framework and technically sound operational workflow to prepare organizations for agentic analytics readiness, featuring a command center, the ability to monitor your agentic analytics assets, and the ability to assign tasks to improve said assets.

This showcase and infusion of the DataFam allows Rekha the opportunity to turn over the stage to a prominent AI-oriented community member, Will Sutton. Will first presents us with the eye-rolling task of servicing a fictitious e-bike company CEO who had the bright idea of including one of the hardest to read chart types ever, a chord diagram, in duplicate, as a bicycle on a dashboard.

But looking beyond the cringe, Will’s demos reveal something much deeper, the most under-realized piece of technology in the room, Tableau’s Model Context Protocol, or simply stated, Tableau MCP. An open-source project turned into an extremely robust protocol for nearly any AI tool you can imagine to access content and analyze data within their platforms. The bright glimmer of AI-fused analytics possibility.

The last section turns out to be the most boring part of the keynote, but also the most cogent version of the Tableau Next workflow process in Salesforce that I’ve ever seen. Salesforce CTO Muralidhar Krishnaprasad, simply known as MK, shows us how an Excel file with two distinct tables is uploaded into Salesforce. Which in turn generates four new artifacts: a semantic model, two data lake objects, and a blank visualization worksheet.

Let me say that back to you. Someone with two tables of data in Excel is being sold that their best course of action is to upload the file into Salesforce and create four new digital assets, just to see a dual-axis line chart of their weekly sales and inventory numbers on one screen.

And while you may argue this is a simplified concept for the demo, the technical layers being thrust upon you prior to actually analyzing the data are made obvious. The technology is not a tool to aid in human seeing and understanding, it is designed for operational control of data.

The close of MK’s Tableau Next demo also marks the end of the three chapters of the keynote. A close that doesn’t come without a “but wait” moment from Southard, featuring a high-concept epilogue built in Figma to show an agentic analytics command center designed to control the whole thing. I can’t help but openly mourn: fingerprints from my hackathon win are faintly legible across this new thing.

So if one thing was on display throughout the keynote, I’d boil it down to this idea: we’ve REALLY REALLY gotta figure out how to organize our data for the AI agents. We seriously need to bring all our data into a system, seriously need to have it organized by technical experts, labeled for business users, and served up in Slack channels.

Aside from the Slack channel requirement, let’s look at the facts. Everything mentioned in terms of the direction of analytics is the same pain that was experienced for years on end. In the last iteration it fell under the dream of self-service analytics, with the most visual metaphor provided by Andy Cotgreave at a prior conference wherein he uses a picture of bicycles messily piled up around racks to convey the notion that analytics platforms needed some guardrails for self-service to work.

And here we are, in the year of the singularity talking about how we need governed, managed spaces for analytics, this time for the chatbots. Because apparently, doing the work for humans, wasn’t sufficient motivation. This is the state of analytics in 2026, not unique to Salesforce, just easier to spot because it’s all in one platform.

At this point the hypocrisy of the software engineers, of the information technology professionals, of the enterprise SaaS nerds is fully on display. Let’s spend our time curating data for the bot army. Tableau’s “help people see and understand data” mission has been hollowed out to nothing. The philosophical underpinnings of what it once meant are gone. Dead.

Instead, it has been replaced with layers. Layers upon layers of abstraction, wherein the meaning of data is touted to be more meaningful because it is sufficiently cleaned, labeled, organized, described, governed, and served.

The entire act of using data to make sense of things, of helping people see more than they saw before, to better understand, has left the building.

Let’s be clear. Cleaner data isn’t more meaningful data, it is just easier to work with. And moreover, it was only difficult because of the mathematical nature of computing. Administrative residue stored as mathematical matrices born out of the way the very technology itself had to be architected.

And so it is time for this 2015 believer to also become the 2026 coroner. A member of the DataFam coming out and saying the uncomfortable truth. What was once the mission that drew me to the conference, Tableau’s, “help people see and understand data,” no longer exists at Tableau.

A statement whose rush of grief leaves me still holding the mission belief, but with no home. An exiled member of the family who must separate herself, forgo the swag, the peer conversations, and the breezy San Diego atmosphere.

What was once a place where I could meet and learn from industry peers, those who valued the human-centric nature of data analysis, that sacred feeling of being in the flow, has become a grave I helped dig.

I would be remiss if I didn’t spend time sharing my practitioner’s observations. Especially given that so many practitioners do go to the conference to learn about new features.

So here, as an appendix, are my top-line notes from the much more technical main stage event, Devs on Stage.

First on my list: a REST API Connector on Cloud — they announced a feature (sorry, I can’t be bothered to know if it is GA, pre-GA, fever-dreamed, or available only on some specific SKU) allowing users to consume data into Tableau Cloud via REST API. This is a great advancement and new feature to the Tableau Classic platform.

By my interpretation, anyone can connect to an application via an API endpoint and turn that into structured data for analysis in Tableau. For the data analyst who wants to bring in data from everywhere for analysis, this is a clear victory.

Remember back when they had Web Data Connectors? Remember when we all clapped that Tableau Desktop could read JSON files? That was in 2015 and 2016 respectively, a decade ago.

They also mentioned published data sources available in Excel. A radically obvious idea that aims to potentially divert Tableau from the narrow lane it unintentionally found itself in: the visualization front end of a long chain of modern analytics tools.

This move plays squarely into what I think Salesforce does best, integrate technologies. They may perhaps have the best group of integration engineers ever. Those whose job it is to take disparate pieces of enterprise technology and build connectors between them.

They also pandered to the Microsoft crowd by way of mentioning Tableau charts as images in PowerPoint. A feature that has been around in varying forms for years.

I remember the sheepish clapping of the crowd when export to PowerPoint was introduced. Now this feature gets better because you can point to the chart object where it exists within your Tableau platform and it will retain the connection. Meaning that if the numbers change in the Tableau chart, they’ll change in your slide deck. This is more of the same, a wrapper-ized API call for the PowerPoint masses. You’ve been able to pull PNGs of Tableau visualizations since pre-API days. The only thing that’s changed is that it is now done via a sidebar menu vs. custom scripting.

The last three that made their way onto my list:

1. Layers Everywhere — rebranding map layers as actual visualization layers. Removing the X,Y coordinate nightmare that many members of the community explored in the pursuit of better ways to visually communicate data.

2. Data Driven Formatting — an unexplained technology demonstration of what practitioners have been doing for years: changing the visual tableau based on the information presented to the end user.

3. Tableau Classic vizzes in Tableau Next — An attempt to create connective tissue between Tableau Next and Tableau Classic by way of omissions and smoothing. The ability to add Tableau visualizations to Salesforce has existed since as long as I can remember. Somehow this has been repackaged and presented as a new innovation, “bring your pixel-perfect Tableau Classic visualizations to Tableau Next with the click of a button.” What’s the point of this? To justify the monopolistic behavior of forcing their customers to pick up CRM SKUs? To force them into a consumption-based pricing model? Certainly it isn’t to bring analytics into the flow of work, because that’s been possible for longer than my cat has been alive.

Those are the only “technical” innovations that merited a line in my notebook, and as you can see from my own commentary, most are rebranded, repackaged, slightly easier versions of what has existed for, in many cases, more than a decade.

And none of them addressing the craft that made Tableau matter in the first place: helping people see and understand data.