Why Dreamforce 2025 will be all about agents, agents, agents

Dreamforce 2025 is imminent, with 50,000 people due to descend on the Moscone Center to hear about the latest advancements across its cloud, data, collaboration, and AI offerings.

In the past few years, Salesforce has gone from strength to strength, with sustained growth and an exploding customer base for its offerings. The CRM giant has gone all in on enterprise AI in moves to meet customer demand, recording 120% revenue growth across its Data Cloud & AI segments in FY25.

CEO Marc Benioff is known for his outspoken views on the technology and last year vowed to “break the hypnosis” built by hyperscalers around AI and unlock real value for businesses.

At Dreamforce 2025, we can expect Salesforce to double down on these goals, with more capabilities and integrations for its existing AI lineup, as well as better integration of the technology across its product ecosystem.

Agents, agents, agents

Salesforce has been a major player in the enterprise AI space for years – with Benioff having voiced enthusiasm for becoming an “AI first company” in 2014 – and Salesforce having launched the first iteration of Einstein in 2016.

But it’s with the widespread adoption of generative AI that Salesforce has put both its feet into AI, first with the OpenAI-powered EinsteinGPT and more recently with the launch of its flagship agentic AI platform Agentforce.

The latter has quickly become a jewel in Salesforce’s crown, with its ability to autonomously complete tasks such as customer service, coding, or filling in forms, based on contextual data drawn across a business’ Salesforce Customer 360.

Salesforce has promised to deliver on what “AI was always meant to be” in the enterprise, with Benioff having pointedly argued that AI ‘hallucinations’ are a major barrier for the technology without the right data and controls.

This has become a key selling point for Salesforce’s own AI platform. Because it operates within Salesforce’s walled garden, it avoids some of the risks associated with common agent protocols – and Benioff has repeatedly criticized competing agentic AI platforms in comparison, dubbing Microsoft Copilot “a huge disaster”.

I recently wrote how, in my opinion, many agentic AI products are overhyped, challenging the idea that agents will automatically produce productivity improvements or growth. In its ultra-practical, closed ecosystem approach to agentic AI thus far, Salesforce is presenting itself as a more surefire provider of AI ROI.

Agentforce’s purpose is to automate and replace specific roles within the Salesforce ecosystem. Take sales support: Salesforce itself has used Agentforce to replace 4,000 sales support staff with agents since the product launched, as bragged by Benioff on a recent podcast.

Expect to hear a lot more about agents and specific advances customers have made with Agentforce this year. New features will likely be rooted in two key updates: first, new proprietary models to succeed Salesforce’s xLAM 8x22b, as well as expanded support for the likes of GPT-5 and upcoming Gemini releases.

Second, and more importantly, Salesforce is likely to make announcements aimed at making Agentforce even more integrated in customer data and even easier to use. When the product was first introduced, the firm promised it wouldn’t burden its customers with ‘DIY AI’ and it could double down on this by reducing the complexity of creating and conversing with agents across its platform.

Slack, which Salesforce acquired for $28bn in 2021, is one channel to watch here. At Salesforce World Tour, held in London in June, the firm demonstrated how users could chat directly to agents within Slack and teased that it could become the go-to front-end for interacting with Agentforce in the near future.

Building on data foundations

Without its firm grip on customer data, none of Salesforce’s AI success would be possible. And while it’s never the star of the show, IT decision-makers know that enterprise data is the real fuel for meaningful improvements.

Slack is also core here, with Salesforce having announced last week that partners and developers will soon be able to connect agents to Slack via a new dedicated API and model context protocol (MCP) server.

The firm says that enterprises will be able to use the vast amounts of unstructured data found in Slack conversations to “enrich AI and agents”

Salesforce’s $8 billion acquisition of Informatica is unlikely to be a prime focus at Dreamforce, as it’s not yet received regulatory approval and isn’t expected to clear until early 2026. But make no mistake: it will be absolutely instrumental to Salesforce’s AI offerings going forward.

One of the main bottlenecks for agentic AI deployment is the data and integrations the tools are capable of accessing. Without the right data infrastructure, businesses simply won’t unlock the level of automation necessary for return on their investment – with the data management and integration Informatica brings to the table, Salesforce could be much better equipped to overcome this issue.

With this in mind, I wouldn’t be at all surprised if some of the announcements at this year’s event are left a little open-ended, with room for the firm to expand on capabilities next year as the extent of its new data integration empire becomes clearer.

ITPro's Rory Bathgate will be covering Dreamforce 2025 live from the Moscone Center, San Francisco from 13 – 16 September. To stay up-to-date with the latest news and announcements from the conference, follow our live blog and subscribe to our newsletter below.