TDX 2026: The 5 Announcements That Will Change How You Deliver Salesforce

2. Agent Script: determinism comes to agents

Here is the problem every CIO has with agents. You type the same prompt twice and you get two different answers. That is fine for a chat app. It is a non-starter for enterprise work.

Agent Script is Salesforce’s answer. It is a flat, versionable file that defines what an agent does, with variables, conditional logic, before-and-after steps, and connections to sub-agents. You can check it into GitHub. You can deploy it like any other piece of code. You can test it. You can roll it back.

Deterministic vs probabilistic: the table your CIO needs

One of the instructors at TDX put it best: “If I’m a C-level executive, I want something that I am going to get the same response back every time. I want reliability.” Agent Script is how you give them that.

Three ways to build an agent now

• •AI-assisted: natural language, fastest to get started, least control
• •Canvas: visual builder, good for refinement
• •Script: the flat file, full control, production-grade

If you are serious about agents in production, your team needs to live in Script view. The other two are for prototyping.

3. The $50M Agent Exchange fund: a new partner economy

AppExchange changed the Salesforce partner world. Agent Exchange is its successor. It is a marketplace for agents, skills and MCP servers. And Salesforce has put $50 million behind it: up to 20 investments for new partners building on the platform.

The $50M Agent Exchange Builders Initiative: up to 20 investments for new founders

Why this matters

If you are an ISV, a consultancy, or a founder sitting on a good agent idea, this is a real route to market. The monetisation plumbing is being built for you.

If you are an enterprise buyer, the calculation shifts. You used to build custom agents because nobody else had them. In 12 months, you will be buying most of the common ones off the shelf. Your own team’s build effort should go into the 10% that actually differentiates you.

4. The Experience Layer: build once, render everywhere

This one might be the most practical of the lot. The Agentforce Experience Layer lets you define a UI once, using a standard schema, and render it natively on any surface: Slack, Teams, Mobile SDK, ChatGPT, Claude, Gemini, your website.

Remember when we had to test everything in ten browsers? Or when mobile meant building a separate app from scratch? This is the same problem. And Salesforce is solving it the same way: abstract the surface, keep the logic.

What changes for delivery teams

Programme scoping gets simpler. Instead of “build for web plus Slack plus mobile” as three parallel streams, you scope one experience and one deployment pattern. Cross-platform bugs become someone else’s problem.

Design reviews get more important, not less. If one schema renders on seven surfaces, getting the schema right is where the quality lives. That is an architect problem, not a front-end problem.

5. Data 360 as the foundation of everything

Data Cloud has been rebranded to Data 360. Fine. That is not the announcement. The announcement is what it now does.

Data 360 is now the foundational data tier for Agentforce, Tableau Next, Marketing Cloud, and a new product called Agentic Enterprise Search. It is where Salesforce logs every agent conversation for observability. It is where unstructured data gets chunked, vectorised, and searched. It is the grounding layer that makes agents accurate rather than hallucinatory.

The three things I was not expecting

• •Zero Copy is now the default pattern. You do not need to move data out of Snowflake, Databricks or BigQuery. You reference it at runtime via query federation, file federation, or caching. Three patterns with different cost and latency trade-offs.
• •All ingestion is now free. Under the new profile-based pricing model, even real-time and accelerated ingestion comes in at zero cost. You pay for unstructured processing and some data processing. Not for bringing data in.
• •Agentic Enterprise Search is coming. It is essentially an out-of-the-box search agent that lives natively inside Salesforce and replaces global search. Search your unified data, structured and unstructured, in natural language. Replaces a lot of custom search builds.

A word of warning from the partner keynote

Data 360 is not an MDM. Do not try to make it one. The speaker was explicit: “Data Cloud is not trying to be an MDM solution. It’s not.” If you already have Reltio or Informatica, Data 360 should receive clean data from them, not replace them.

What it all means for your next Salesforce programme

I have been to a lot of these conferences. TDX 2026 felt different. The announcements fit together. Headless 360 exposes the platform. Agent Script makes agents reliable. Agent Exchange creates the marketplace. Experience Layer solves distribution. Data 360 makes the whole thing accurate.

If you are scoping a Salesforce programme right now, here is what I would do differently on the back of this.

• •Stop designing UI for specific surfaces. Design the experience once. Let the Experience Layer handle the rest.
• •Check Agent Exchange before you build. Every agent you do not build is six months you can spend on something that actually differentiates.
• •Invest in Agent Script skills in your team. Not Canvas. Script. It is where production work lives.
• •Treat Data 360 as the grounding tier, not a CRM add-on. Your agents are only as accurate as the data they can reach.
• •Get your integration architect in the room early. Four integration patterns now, not one. The wrong choice costs.