Another Dreamforce Is In The Books: Here's What's Actually Worth Paying Attention to

Dreamforce has always been Salesforce's biggest stage for AI news, but this year felt like an inflection point rather than another incremental Agentforce update. The event's official theme, "The Agentic Enterprise," centered almost every marquee announcement on getting more value out of AI agents rather than showcasing the core Customer 360 applications. Here's what we think actually matters for organizations running Salesforce.

1. AIforce: a new interface layer above everything

The single biggest announcement of the week was AIforce, unveiled by CEO Marc Benioff as a "live interface layer" positioned above Agentforce, Data 360, and Customer 360. Rather than another point product, Salesforce is framing it as the layer that brings the full power of the platform to wherever people and agents actually work, turning the existing permission model into the security boundary for every new surface it touches. AIforce launched with three pre-configured interfaces: Claudeforce, Slackforce, and Agentforce Coworker.

2. Claudeforce moves the Anthropic partnership from roadmap to product

This is the one we've been watching closely. Claudeforce is a plug-in that brings Salesforce directly into Claude, and on the keynote stage Salesforce announced that its Sales Cloud skill for Claude is now in open beta and free to use, demonstrating the Anthropic partnership live. For organizations already exploring Claude for day-to-day work, this closes a real gap: Salesforce data and actions become reachable without leaving the chat interface. It's also the clearest sign yet that the model layer is opening up, not staying locked to one vendor. On the infrastructure side, Agentforce customers now have access to Amazon Bedrock's full range of frontier models, including Anthropic and NVIDIA, with OpenAI models coming soon, and Google Cloud's Gemini models now sit behind Agentforce's Prompt Builder and Reasoning Engine.

3. Koa: Salesforce builds its own reasoning model

Koa is Salesforce's first CRM-specific reasoning model, post-trained on NVIDIA's Nemotron 3 Super. The pitch is that reasoning becomes a swappable component inside the platform rather than something every agent has to be built around individually, which matters for governance as much as for capability.

4. From "build your own agent" to pre-configured digital workers

A real strategic shift: Salesforce introduced a family of named, job-ready agents on top of Agentforce, including Casey, Paige, Carter, Hunter, Marshall, Piper, and Fin, with all but Hunter (still in pilot) available now. This is a departure from the last two years of messaging, which leaned heavily on customers building custom agents from scratch. Pre-configured, role-specific agents are now the default entry point, which should meaningfully lower the lift for mid-market teams that don't have deep in-house AI engineering.

5. The "Enterprise AI Harness" ties it all together

Salesforce described this as the architecture bringing together everything agents need to understand the business, reason and plan, take action, and operate within enterprise controls, spanning context, agency, action, governance, security, and models through one composable architecture. Every other announcement of the week effectively sits inside this framework, it's the connective tissue between AIforce, Koa, and the new agents.

6. The ROI conversation finally has data behind it

For the last two Dreamforces, the honest criticism was that Agentforce adoption numbers were thin. That's changing. Salesforce's own Agentic Enterprise Index, built on five quarters of production activity across 400 companies and a survey of nearly 5,000 professionals, shows active agents per organization growing from 5 to 13, average time to put an agent into production dropping from 4 days to 1.9, and 7 in 10 customer service sessions now closed by an autonomous agent rather than a human. The Agentforce keynote itself was explicitly framed around moving from fast start to real ROI, which tracks with what we're hearing from clients: the pilot phase is ending.

7. What it means for admins

The Admin Keynote focused on where the admin role is headed as Agentforce and the Headless 360 platform change how work gets built, with an emphasis on the security and governance tooling teams need to deploy AI responsibly. Admins aren't being displaced by this shift, but the job is clearly moving from configuration toward agent oversight and governance.

The takeaway for Salesforce customers: this was the year Salesforce stopped selling "agents" as a concept and started selling infrastructure to run them at scale, with real model choice (Claude, Gemini, Bedrock's catalog) and real adoption data behind it. For teams still in pilot mode, the gap between "experimenting with Agentforce" and "running it in production" just got a lot more defined.

If your organization is still in the "we're testing Agentforce" phase, Dreamforce 2026 was the clearest signal yet that the window for calling it a pilot is closing. As a Salesforce partner since 2013 and a member of Anthropic's Claude Partner Network, we're already helping clients figure out where Claudeforce, the new agent family, and the Enterprise AI Harness actually fit their roadmap, not just the demo. If you want to talk through what any of this means for your org specifically, reach out and we'll walk through it together.

‍