This year’s Dreamforce is unlike anything else. The CEO and co-founder of Salesforce, Marc Benioff is on fire. He is a lot more dominant this year, outspoken, demanding, but above all driven. Driven to present Agentforce and get the message across that Salesforce is doing something special with AI. Microsoft also gets its ass handed to it.
At most IT companies, you can rarely catch the CEO making a bold statement. From the organizations, all communication, every speech, keynote and presentation is written out in advance. Every word is put on a scale, which then results in a neat and correct story. Nothing wrong with that in principle, everything is between the lines, but sometimes it also gets a little boring.
Marc Benioff, the CEO of Salesforce, usually stays between the lines, but not this year. Not only is he passionate about the new Salesforce solutions, he is also outspoken about the competition. At Salesforce, they have taken a big step forward with AI. For large organizations, Salesforce has now positioned itself as the AI platform. No complex pathways and building your own AI, just roll it out in a few clicks.
Benioff seems a different person, shaken up, and perhaps his ego got a hit last year
It is refreshing, though, that Benioff is so outspoken. No longer is everything politically correct, the jokes he makes are suggestive enough. The answers he gives don’t lie, either. A journalist from Bloomberg who compared Salesforce to Microsoft was decisively put in his place. There is new energy in the CEO of Salesforce, he is driven and demanding toward his staff. That’s very clear.
It’s possible Benioff was shaken up last year with the arrival of an activist shareholder who suddenly stepped in at Salesforce and demanded significant changes. Salesforce needed a major reorganization, while Benioff always talks about ohana (family), when it comes to Salesforce people. Ultimately, Salesforce got the activist shareholder out again without too much consequence. But it probably did shake him up and perhaps put a dent in his ego.
After 25 years of leading Salesforce, he may have been slightly less sharp and demanding than in the early years. The activist shareholder might have caused the fire to return at the Salesforce CEO to strive for innovation more than in other years. Benioff almost lost control of his own company, which sounds like a huge wake-up call.
Innovation is certainly present this year, as Salesforce is making major strides in AI. Salesforce allows organizations to get started with AI quickly without having to experiment and build everything themselves.
“Stop building your own AI, it doesn’t work.”
According to Benioff, we are now done with DIY AI, or “do it yourself” AI, as it rarely works. Many organizations have tried to build AI solutions with the well-known big LLMs, throwing a huge amount of data at it and hoping that a good result will roll out. However, the fact is that this does not work well enough, the AI goes hallucinating too often, the result is not reliable enough. Internally, it could still be used, as an employee could then decide if the AI result is useful. Opening the AI to customer-facing communication is too risky. That is why most SaaS vendors choose to only make suggestions based on the LLM’s.
AI works well within a clear context, use case and with high-quality data
To make AI work, you must make a smaller use case, with a clear context and high-quality data. However, you also need a process that can monitor, verify, and if necessary, rerun the AI for a different output. That is very difficult for organizations to develop on their own. A standard LLM like the ones from Anthropic and OpenAI are designed for generic use. You can use those to generate a very specified text, but if you give it too much data, or let it be creative with large data sets, things quickly go wrong. You have to think about the context and which data you feed them, you can’t simply give them everything.
Or as Benioff puts it:
“CoPilot is the next Clippy.”
Time will tell if Microsoft CoPilot will fail, but Microsoft is struggling to get it working properly. Adoption among organizations is also lagging because it is not compelling. Indeed, there is a chance that for very broad use cases, it will end up like Clippy.
Tying a lot of enterprise data to an OpenAI LLM and having the user generate outcomes based on that data doesn’t deliver enough quality. The context is too broad, and the purpose is too unclear. Add to that the fact that it is also very costly, and you have a case that cannot succeed.
AI from a platform has more potential
The difference with Salesforce is that based on the various SaaS applications they can add a lot more context to an AI. The future of AI, according to Benioff, lies in AI agents, or whats Salesforce calls: Agentforce. Agents are virtual bots that act like humans and have very distinct skill sets and actions with which they can serve a user. Salesforce customers can even build AI bots without programming (no code) with a few mouse clicks and deploy them to various Salesforce applications. These AI bots work through a chat system, WhatsApp, iMessage, and by voice over the phone.
Salesforce Data Cloud is the foundation of Agentforce
In recent years, Salesforce has made a significant push on its Salesforce Data Cloud. This contains all Salesforce data, metadata, and context stored from all applications on the Salesforce platform. In addition, the Salesforce Data Cloud can interface with external data sources such as Snowflake, Databricks, and BigQuery, giving Salesforce access to all corporate data.
From the specific context of Salesforce applications, such as sales, service, marketing, commerce, and so on, AI bots are developed within a clear context. Such an AI bot can be equipped with various skills and flows that many customers have already created on the platform.
Salesforce has also built a so-called Atlas Reasoning Engine, which is the link between the data cloud and the AI agents. Atlas checks each step of the AI agent and verifies the outcomes. If they deviate even slightly, Atlas intervenes. As a result, according to Benioff, Agentforce is enormously accurate, rarely, if ever, derailing or hallucinating. This puts Salesforce ahead of the rest of the market.
Salesforce is accelerating and coming to you this fall
Of course, Salesforce has yet to prove itself, something we will explore further this week. In any case, Salesforce will host about 25 more significant events in different countries worldwide in the coming months. The goal of those events is to help local customers develop AI agents. Benioff states that 1,000 customers should have developed AI agents this week. Thousands more should be added this fall, and at Dreamforce next year, 1 billion people worldwide should have used a Salesforce AI agent. An impressive goal.