Forget the copilot: Salesforce is appropriating the AI agent. This iteration of AI should represent a new wave of innovation, and will become ubiquitous. However, the idea of AI agents predates Dreamforce and is already certainly not unique to Salesforce. How important are agents really?
The era of the AI agent has arrived. So says not only Salesforce CEO Marc Benioff, but also Sam Altman of OpenAI. Altman recently described “level-3 agents,” which can reason and may be trusted with a degree of autonomy. The AI guru’s description fits perfectly with the agents Benioff envisions.
Earlier, Google launched the Vertex AI Agent Builder, which allows companies to build AI agents, and Microsoft is also a big believer in AI agents. Hearing Benioff and Clara Shih, CEO of Salesforce AI, busy during Dreamforce, would almost forget that the idea of the AI agent is not new. Presented with an almost aggressive enthusiasm, Agentforce seems like the only solution.
Trumps and confusion
Salesforce has assets. Agentforce integrates with the existing Salesforce environment, can access enterprise data via Data Lake, and is compatible with existing Flows and APIs. In other words, with Agentforce, Salesforce lowers the barrier to deploying an AI agent.
For the last eighteen months, the world was confused
Clara Shih, CEO Salesforce AI
Shih likes to go a little further. According to her, Salesforce is the only company that has all the building blocks in-house to roll out a full-fledged agent. The rest of the world needs to be educated. “For the last 18 months, the world was confused,” she says. “People thought that because they could just use ChatGPT, they could easily bring such a model to an enterprise context. But that’s not how it works.”
Sense of exaggeration
Only: no one really thought so. Microsoft, OpenAI and Google marketed Copilots, which can support employees using company data. Making an AI copilot work within defined boundaries proved difficult. For example, the Microsoft Copilot contained a bug that allowed sensitive data to be accessed anyway.
This is painful, but also somewhat inevitable with new technology. Who says Salesforce agents will never know a vulnerability? Disney, which Marc Benioff says he is an immense fan of, announced last week that it was moving away from Slack (part of Salesforce) because of security concerns. Benioff himself does not speak of a security problem, pointing the finger in an interview to Bloomberg to phishing and social engineering as the reason for the data breach. There is no concrete security report surrounding the data breach (yet).
That other data cloud…
Snowflake also drew extensively on AI at its conference. There we again saw systems aimed at business use, with models reliably and securely providing correct answers based on business data. Snowflake even developed its own efficient retrieval augmented generation (RAG) model that at first glance does bear some similarities to the Atlas Reasoning Engine that powers Agentforce.
Snowflake, moreover, also speaks of fundamental pillars for enterprise AI. Of course, according to that founder Benoit Dageville , Salesforce is not the place to be for that.
Cards on the table
For Dreamforce 2024, a lot of cards were already on the table:
- The idea of autonomous agents has been around for some time. Microsoft, Google and OpenAI were already working on it, as was Salesforce. We even saw early implementations, such as the Vertex AI Agent Builder.
- The importance of security was never in question. Employees may have thought they could inject corporate data into ChatGPT, but companies large and small had long been tinkering with ways to integrate AI securely.
- The right data is critical, Snowflake told the same Moscone Center in San Francisco a few months before Salesforce. Like Salesforce, Snowflake argued that the solution is to build AI on top of the Data Cloud. Only: Snowflake had a data cloud before Salesforce.
So Agentforce is not as revolutionary as the Benioff-Ohana would have everyone believe. On the other hand, Agentforce is indeed more than a marketing term. Salesforce effectively has something that no one else can offer right now.
iPhone moment?
Customers who have fully embraced the Salesforce platform have the foundation for agents all set. No other party can currently offer a similar out-of-the-box experience that is even SME-friendly. Salesforce owns the data, the automations, the internal and external channels, the API integrations, and can now click agents between them. That’s powerful.
read also
No Clippy, but Sophie: AI according to Salesforce passes the (Microsoft) Copilot
After all, inventions that change the world are rarely true inventions. Steve Jobs did not invent the smartphone: numerous smart phones existed before the iPhone. In fact, the PDA dates back to the 1990s. However, Apple merged all the relevant components into one user-friendly whole.
In fact, Salesforce is now doing the same with agents. Are we experiencing Benioff’s iPhone moment? In any case, Shih believes agents will become as ubiquitous as apps and Web sites in the coming years.
Not for everyone
However, Salesforce is missing one thing: accessibility. Anyone could buy an iPhone in 2007. You didn’t have to own a Mac computer as a prerequisite. It’s not like that with Agentforce. If you want to use Salesforce’s convenient, quickly deployable and secure agents, you need Salesforce Foundations. You have to be a customer and have your data in the Salesforce platform and Salesforce Data Cloud, or you can’t get started.
Salesforce may be able to accelerate agent adoption. The agentforce announcement is especially relevant for those already on board with the Salesforce story, but will nevertheless force competitors to think again. How can companies outside Salesforce get started with agents quickly and securely? And how do you democratize AI agents for companies large and small? Can you safely integrate AI agents at all without going all-in on a single ecosystem?
Benioff and Shih presented that challenge to Microsoft, OpenAI and the rest. Talking about agents and experimenting with them is one thing, rolling them out at scale another. Even if that scale is limited to their own customers, the competition still has to come up with a good answer.