On September 17, Vantegrate streamed its Thursday masterclass live from Dreamforce in San Francisco. This time there was no classroom: there was floor 61 of the Salesforce Tower, hallway noise and a special guest.
The full class is on YouTube. Watch the replay to hear the answers firsthand.
Juan Manuel (JuanMa) hosted alongside Francisco (Fran) and Josué. The guest was Alba Rivas, who works at Salesforce and had just come off the stage of the developer keynote.
Adrián Barbani, architect of the Latin American ecosystem, and Gustavo, who leads operations for Latin America, also stopped by. From Buenos Aires, Angie and Ceci joined in, reading the questions from the chat.
The central topic was making sense of the noise from the announcements. What AIforce is, what happens with Agentforce and how the role of the developer changes when agents write a good part of the code.
The context: a masterclass from floor 61
JuanMa kicked off with a tour of the famous Ohana floor, floor 61 of the tower. It is circular, with a view of the whole city and (so they say) free food all the time.
The Vantegrate team had just given six talks at the event. They started with four planned and ended up adding two more.
AIforce: what it is and what it is not
Fran put it bluntly: with so many names and renames, he was confused. He asked Alba to sort out what is new, what already existed and what overlaps with what.
Alba opened with humor. Salesforce wouldn't be Salesforce without a rebrand, but she clarified that AIforce is not a rebrand of Agentforce: it is a broader concept.
From Headless 360 to Headless Toolkit
To understand AIforce you have to go one step back. At TrailblazerDX (TDX), people started talking about Headless 360.
The logic was this. Salesforce detected that customers want to work from where they already are, and that there are more and more interfaces that are not the platform's classic UI.
Headless 360 was the set of technologies that makes that possible. Alba split it into three pieces:
MCP Servers: they let you expose business logic to any agent.
Headless Experience Layer: in addition to text, it creates components and interfaces that appear dynamically inside agents.
Multi framework: you no longer have to use only Lightning Web Components to build on the platform. It can be Angular, React or another framework.
That package is now called Headless Toolkit. The technologies are the same; only the name changed.
The layer that opens Salesforce to any agent
AIforce comes in on top of that foundation. Alba described it as the layer that uses all those technologies and opens Salesforce to any MCP client or agent we know of.
It also adds partnerships and products with agent vendors. The goal is to make setup as easy as possible and data access secure.
The keynote had many guest CEOs (Anthropic and NVIDIA, among others). One example is Claudeforce, which quickly configures Claude to work with Salesforce, and specific agents are added on top of that.
Agentforce is not going away
Here came one of the most valuable clarifications of the afternoon. The idea that AIforce replaces Agentforce is, according to Alba, a misconception that may have been poorly communicated.
Agentforce is still just as important. JuanMa summed it up as Salesforce's own AI, and Alba confirmed it.
If you are building agents inside the platform, none of that becomes outdated. You can review the approach to AI agents to see where each piece fits.
Agentforce Coworkers and semantic search
Another common confusion is Agentforce Coworkers. Alba explained it as an agent that comes ready-made and knows everything about your organization.
You can ask it the questions you used to throw at search and that returned nothing. It works because it has semantic knowledge of what is in the org.
Alba said she gave a talk on Salesforce Catalog. That product categorizes all of the org's metadata with semantic information.
Thanks to those layers, agents search by meaning and not only by the exact word. They find synonyms and similar terms.
How to explain a semantic layer to a business user
JuanMa asked for the version for a middle manager. Alba's answer was very clear: they can ask about data without knowing exactly what it is called.
The agent searches by similarity for the best match within the platform. That said, the result depends heavily on how the underlying system is implemented.
Skills in Markdown
About the Coworkers skills, Alba was honest: she hadn't tried them. She inferred they would be custom instructions about the agent's behavior and about how to run searches.
JuanMa confirmed the inference was exactly right. Skills are written in Markdown, a text format very natural for an LLM, and they can be shared among coworkers.
The new role of the developer
Alba's talk at the developer keynote had three chapters. How the developer's role changes, what role AIforce plays and how to create headless experiences that appear in external agents.
For the first chapter they invited a teammate who was one of the first to adopt AI. They asked him to show in 15 minutes how he works today, with an enormous number of skills configured.
From code to guidelines
Alba described the classic work. Create an object, build the queries, write the class, write the tests, deploy and see whether it works.
According to her, that work doesn't have much time left. The developer's role is going to shift and, at the same time, expand.
The developer now accompanies the agents at several points of the development cycle:
Design: ask the agent for a solution that follows your guidelines, technologies and criteria.
Review: check that what was built matches what was requested, because AI also makes mistakes.
Guardrails: give the agent knowledge and set limits on what it can do.
Can everyone develop?
Someone had asked Alba that before the livestream.
Yes, anyone can ask an agent to build an application. No, not anyone can build an application that works in production in an enterprise environment.
If you work in the Salesforce for developers ecosystem, that is the difference that defines professional value today. It is not about writing lines of code; it is about making sure what reaches production works.
Roles that blend together
Fran asked where the role is heading. More architecture, more business, something hybrid?
Alba thinks it is a mix. She said that within Salesforce the Admin Advocates team merged with the Architect Advocates team, because the line between those roles is getting thinner.
The developer who only writes code will no longer be needed. What will be needed is reviewing that code, giving guidelines to agents and making sure it is well integrated into the overall architecture.
The business view: processes, people and technology
Gustavo brought the business perspective. For him, the message at Dreamforce was about growth, tied to the transformation of companies.
That transformation has three keys: processes, people and technology. What matters is that they work together and that the people who know the business are the ones who drive it.
What struck him most was a moment in the keynote. Benioff brought people on stage to show where Salesforce stands: it is the one that manages the data.
When Fran asked about the future of roles, Gustavo did not hesitate. Roles will unify and everything will point toward the business.
The return of the business analyst
JuanMa asked Alba whether we are seeing a resurgence of the business analyst. She confirmed that this role (and the program manager role) is gaining a lot of weight.
The reason is concrete. Agents need specifications: the developer writes the technical one and the business analyst writes the business one.
The better the specification, the better the agent works. That is why it matters to understand each customer's processes and to have the human skills to listen to different people.
If you want to apply that conversation to your industry, the Salesforce page has the map of what we do with the platform.
Is Salesforce marrying Claude?
Adrián threw out the opinion question before leaving for another talk. Before Dreamforce, Claudeforce was announced and it looked like a marriage with Claude, but OpenAI and NVIDIA also spoke at the event.
Alba answered that the relationship with Claude is moving forward. The Salesforce plugin for Claude has already been released for developers.
The Trust Boundary
Alba recalled something that has existed for a while. The Anthropic models called from Salesforce (Agents, Prompt Builder, Model Builder) work under agreements so that company information never leaves.
Salesforce calls it the Trust Boundary, the platform's trust limit. That part of the partnership is already done.
But AIforce is with everyone. The idea is to make it easy for each person to work from wherever they need, without being tied to a single provider.
JuanMa read it this way: MCP is becoming a universal protocol so that everyone can communicate with everyone. Topics like security and data transparency become the center of that openness.
The Dreamforce of MCPs and skills
Josué summed up his impression of the event. For him, this was the Dreamforce of MCPs and skills, because agents advanced very fast and now it was time to organize everything around them.
Alba agreed. It is about openness, and not only for developers: business users also use MCP.
Data 360 configured with an agent
Alba explained how she built the demo for her Data 360 session. She used the Data 360 MCP Server and, with an agent, created her DMOs, calculated insights, transforms and search indexes.
She used it to configure and make the most of the platform, not to code. For example, running those calculated insights and viewing customer results.
Tableau Next and dynamic interfaces
She also mentioned the Tableau Next MCP. What used to require opening the application and applying filters is now just a question to an agent.
The agent returns the chart you need in real time, and you can interact with those components. If you are interested in conversational BI, see how we approach it in Metrix.
Data 360 and MuleSoft: do they cannibalize each other?
Josué asked whether Data 360 is going to eat MuleSoft, after seeing demos with connectors that bring in data directly. Alba thinks not.
Data 360 is there for scalability. Her Catalog session was built there because customer configurations are so large that they cannot live in the core.
Agent interaction logs, governance and observability are also stored in Data 360. It is Salesforce's scalable database.
MuleSoft, on the other hand, handles integrations at the business logic level. That is where Agent Fabric comes in.
What Agent Fabric is
Alba clarified that it is not her area, but she gave a summary. It is a platform within the MuleSoft world that lets you connect agents to each other so they can communicate.
The comparison she used was several agents running in parallel on a computer and talking to each other. For her, Data 360 and MuleSoft serve different purposes.
The new voice of Agentforce
JuanMa saw the new Agentforce voice at the Moscone Center. It sounds human, with breaths, so a conversation with a customer doesn't feel robotic.
Is all interaction going to be by voice? Alba thinks it will be complementary.
She compared it to blogs and videos. She prefers to read, others learn better by listening, and each person will work through the channel that suits them best.
Credits, consumption and Agentforce Vibes
Angie brought a question from the chat. How is the market reacting to the credit-based consumption model?
Alba didn't want to guess. She said something had to be done, because if the ecosystem changes this much, Salesforce has to monetize what it builds.
JuanMa gave an example of that. Two years ago the price was two dollars per conversation with Agentforce, the market didn't like it, and it moved to Flex Credits.
Alba added another case. In Agentforce Vibes they had removed the free portion, and at Dreamforce they announced that a free tier is coming back with a portion of consumption at no cost.
If you're evaluating consumption and licenses, on our pricing page you can see how we approach it at Vantegrate.
AI-generated code: quality and security
JuanMa asked her about the risks of AI-generated code. Two fronts: quality and security.
On quality, the main risk is not supervising what the AI does. The way to improve it is to have the system well set up, with clear skills and guidelines, and always with review on top.
On security, the message was just as firm. Many people believe they can build an app, put it into production and not know anything, and that is not the case today.
The number one piece of advice
Her final recommendation was direct: if you're not using AI, start now. She says she has become much more efficient.
Tasks that used to take her hours she now solves with AI, and in parallel too, with several agents working at the same time.
Alba and her history with Dreamforce
Fran took her away from the technical side for a while. Alba said she first came thanks to a partner, Certinia, which paid for her trip if she spoke about the company.
Once she started working at Salesforce, everything changed. Dreamforce and TrailblazerDX are the two work peaks of the year, and preparation starts months in advance.
That day she was giving her second keynote. She explained that a keynote is the most important talk, on stages for about two thousand people, and more visionary than a regular technical talk.
A community that doesn't stop
JuanMa recalled that the project started with Fran and today has more than 2,000 students every Thursday. Everything is free.
One Thursday there's English for Salesforce with the teacher, and the next one is a masterclass with the architects Josué, Francisco and Esteban (Esti). Alba left with an open invitation to come back.
If you want to follow the thread, on Pulse we publish the material from each masterclass. In resources you'll find downloadable content to go deeper.
And if you want to see how to apply AIforce, Agentforce and MCP in your operation, you can book a demo with the team. The question is no longer whether agents are coming, but who gives them the right guidelines.
