Scoring the Demos Against Sunday's Prediction

A line-by-line look at what shipped, what's roadmap, what's vaporware, and how the predictions from Sunday's pre-keynote post actually held up.

On Sunday we published a pre-keynote post predicting that TC26 would follow the Salesforce Headless 360 playbook: existing API capability, repackaged as agentic innovation, marketed under new vocabulary, and pushed toward customers who would be told to upgrade to access it. That post laid out the technical foundation in detail, including the Hyper API support for multi-table extracts since 2021.4, the wrapper architecture of Tableau MCP, and Tableau's own marketing copy admitting that Tableau Semantics is a layer over data investments customers already own.

Tuesday's 68-minute opening keynote ran. We watched it. The prediction held. In two specific ways, it held even more strongly than we anticipated.

This post is the scorecard. It does not re-litigate what Sunday's post already established. It scores the eight specific announcements against the existing release notes and public APIs, surfaces the moments where the keynote went further than we predicted, and addresses the strongest defense we have heard from Tableau-aligned readers in the days since.

The Eight Demos, Scored

The keynote organized its content around three pillars (knowledge, decisions, actions) and showcased eight specific capabilities. Here is the scorecard.

1. Composable data sources. Tableau presented this as an unveiling. "What we are now announcing is something called composable data sources, where you can take different data, different published data sources, and make them as one." This is the same feature announced at TC24 in May 2024 and underpinned by Hyper API capability since 2021.4, as we documented Sunday. Verdict: Recycled announcement. The capability has been GA for almost two years.

2. Conversational analytics in Tableau Cloud and Server. Multiple presenters described this as a flagship innovation. The quiet part: Mark Recker, the new GM, said directly near the end of the keynote, "It's coming in July to cloud and then server a little bit in the fall." So this is a roadmap commitment, not a release. Customers watching the demo and assuming they could open Tableau Cloud Tuesday afternoon and start using it will be disappointed. The capability is not shipping for at least two months in Cloud, and four to six months in Server. The underlying conversational interface, including natural-language follow-up over published data sources, has been a Tableau Pulse capability since February 2024. Verdict: A UI extension of two-year-old Pulse capability, not yet shipping.

3. The analytical knowledge graph. "We are offering to all of you a analytical knowledge graph. This connects to all your data. It understands it, the semantics, the maps. Don't worry, you don't have to build this. We're building this for you automatically." This is Tableau Semantics, the layer Tableau built on top of customers' already-published data sources, organized into a graph structure an LLM can traverse. We covered the marketing copy admission ("maximize your existing data investments without migration") on Sunday. Verdict: Existing semantic metadata, repackaged as a knowledge graph because that is the term LLM marketing uses in 2026.

4. AI-assisted semantic model building in Tableau Next. "Call it AI to build your AI." This is the most defensible "new" capability in the keynote, and even here the novelty is shallow. Tableau Agent, originally Einstein Copilot for Tableau, has been GA since June 2024. The TC26 announcement extends agent assistance to semantic model construction specifically. The category itself is competitive: Snowflake's Cortex Analyst, Databricks' Unity Catalog AI features, and Sigma's natural-language model building have all been shipping equivalent functionality for over a year. Verdict: A genuine extension of existing AI-assist functionality, but a feature among many in a competitive landscape, not a category-defining innovation.

5. Pulse insights and recommendations in Slack. Demoed as part of the new agentic flow-of-work experience. The CEO of the fictional bike company asks Slack a question, gets an answer, gets a recommendation, and produces a report. Tableau Pulse Slack Digest has been generally available since February 2024. The Slack-native conversational follow-up, the contextual insights, and the natural-language summaries have all been Pulse capabilities for over two years. Verdict: Two-year-old GA capability, demoed as if it were the new agentic flow.

6. Unified structured plus unstructured data. The presenter uploaded an unstructured analyst report and integrated it with structured sales and inventory data, with the agent performing root cause analysis across both. This is the most genuinely interesting demo of the keynote. It is also listed in the April 2026 release notes as Beta in Tableau Next. Beta means not GA, Tableau Next means not Cloud or Server, and "in Tableau Next" means it requires the new product tier customers will be asked to license separately. The underlying retrieval-augmented-generation pattern has been shipping at Snowflake, Databricks, and across the OpenAI ecosystem for over a year. Verdict: Real capability, useful, but Beta-only, locked behind Tableau Next licensing, and behind where the broader market has been for a year.

7. Tableau agents taking automated actions across enterprise systems. "Tableau is no longer just a passive dashboard anymore. You can now connect them to all the actions in your enterprise. Now you can actually use the same Tableau agents with all the actions between with our Agentforce." This is Agentforce, which launched in September 2024. The actions being executed in the demo (rebalancing inventory, dispatching shipments) are Agentforce actions calling Salesforce flows, not Tableau capabilities. Tableau is being wired in as a data source and a surface, not extending its own platform. Verdict: Agentforce, with Tableau as a connected surface. Useful for Salesforce customers. Not new and not Tableau-led.

8. The Agentic Analytics Command Center. A vision for a unified control plane to monitor agent performance, adoption, trust, and data integrity. This was the most honest moment of the keynote. Recker said it directly: "This is a vision demo. None of this is baked. Literally created in the last couple of weeks. It's our idea. It's our vision. We want your feedback." To the team's credit, they labeled this clearly. To customers' detriment, the rest of the keynote did not extend the same clarity to features that are also not yet shipping. Verdict: Vaporware, openly acknowledged. The feature does not exist.

Two Things That Went Further Than We Predicted

The Sunday post argued that the keynote would be a re-announcement of existing capability under new vocabulary. Two specific dimensions of the keynote went further than that.

Three of the eight headline demos are not actually shipping. Conversational analytics in Cloud and Server is roadmap. Unified structured-plus-unstructured analytics is Beta in Tableau Next. The Command Center is a vision deck assembled in the last two weeks. Customers watching the keynote have no way of knowing, from the framing alone, which features are GA today, which arrive months from now, which require migration to Tableau Next, and which do not exist. The opacity is the point. The marketing trades on the ambiguity.

The vibe coding demo deserves its own note. A community member used Claude to write a Tableau Viz Extension based on a hand-drawn sketch, then used Claude again to embed that extension via the Tableau MCP server. The framing was that Tableau MCP enables this. The reality is that Claude's general code-generation capability, applied to the existing Viz Extensions API that shipped in 2024.2, produced the result. Anyone with Claude access can do this for any platform with a public API. Demoing Claude's capability on the keynote stage and labeling it Tableau innovation is exactly the pattern of borrowing other companies' work and presenting it under your own brand.

"But the GUI Is the New Part"

The strongest defense of the keynote, and one we have heard repeatedly in the days since the pre-keynote post went out, is that the API has always been able to do these things, but the productized graphical interface is genuinely new. A native UI for composable data sources, a click-through experience for conversational analytics, an integrated semantic model builder. Code is hard. Most business users do not want to touch code.

The technical part of this argument is correct. Tableau's VizQL engine, the query layer that resolves multi-fact relationships and composable data sources at runtime, is genuinely the part that customers cannot rebuild themselves. Nobody is recreating VizQL on top of a public API. A packaged GUI on top of an existing API has real value.

The argument fails for three reasons.

First, "we built a GUI on top of capabilities our APIs already exposed" is not what was announced from the keynote stage. The framing was agentic transformation, the world's first agentic analytics platform, a new era of decisions and actions. A multi-billion-dollar company unveiling a UI for capability its APIs supported in 2021 is not a category-defining moment. It is overdue product work, marketed as innovation.

Second, the timeline is the tell. Composable data sources were announced at TC24, eighteen months before this keynote. The underlying Hyper API support shipped in 2021.4, almost five years before this keynote. If a productized GUI were the genuine deliverable, it could have been built and shipped in a single release cycle. As we documented Sunday, we built a working version of that GUI in 2024 on the public APIs and pitched it to Tableau. There was no interest. The decision not to ship was not a technical limitation. It was an executive prioritization choice.

Third, even granting that the productized GUI is the deliverable, Tableau is not the only or even the leading place to get a productized GUI for composable data, conversational analytics, or AI-native semantic modeling. Sigma, Snowflake, Databricks, and others have been shipping equivalent or better packaged experiences across the same window. The "GUI is the new part" defense quietly concedes that Tableau is competing on UI delivery against companies that have been delivering UIs at a faster cadence with stronger underlying engines for years.

What Was Conspicuously Not Announced

A list of what we did not see in this keynote is, in some ways, more revealing than the list of what we did.

There was no announcement of new Hyper engine capability. No new connection types. No new performance benchmarks for VizQL. No deprecation timeline for legacy features. No new pricing transparency. No discussion of what existing Tableau Cloud and Server customers get versus what requires upgrading to Tableau Next. No partner ecosystem expansion. No new SDK or developer-facing primitive that was not already in the April 2026 release.

What we did see were two years of existing capabilities, demoed in rapid succession against a fictional bike company, framed as a new era. The capability layer was static. The vocabulary layer was fresh.

What Sunday's Post Said, Confirmed

The recommendations from Sunday stand. Read the API documentation, not the keynote slides. Ask your account team what specifically is changing in Tableau Server, Tableau Cloud, Tableau Desktop, and Tableau Prep that you cannot already do today, and ignore any answer that requires migration to Tableau Next as a licensing motion rather than a capability one. Recognize that the partner ecosystem and the broader modern data stack have been shipping these capabilities for years.

What Tuesday added is the receipt that the pattern is now self-evident. A keynote that re-announces two-year-old features, demos vaporware as if it were product, and ships three of its eight headline demos as roadmap or Beta is not a company in a moment of transformation. It is a company performing one.

When the keynote ends and customers go back to their day jobs, the work of getting actual analytical value out of their data continues. That work is increasingly happening on stacks that ship capability rather than vocabulary. Tuesday only sharpened the case.