Day 1 of Dreamforce 2026 introduced Salesforce’s vision for a world where CRM can meet employees wherever they work.

Day 2 started answering the more difficult question:

What does that actually look like inside a business?

Across keynotes covering Sales, Agentforce, Revenue Cloud, AIforce, architecture, and Data 360, Salesforce moved beyond the broader AI vision and spent much more time demonstrating how agents could fit into real business processes.

The recurring themes were measurable ROI, automated CRM, trusted business context, human oversight, and agents that can work across increasingly complex processes.

For Revenue Operations leaders, Day 2 may ultimately be more important than Day 1.

The technology is becoming easier to understand.

Now organizations have to determine where it actually belongs.

Sales Cloud Is Moving Toward a CRM That Works for the Seller

One of the biggest themes from the morning Sales keynote was a fundamental change in how Salesforce describes CRM.

Historically, CRM has largely depended on people recording what already happened.

A seller has a meeting. They update the opportunity.

They speak with a prospect. They log the activity.

Something changes. Someone updates Salesforce.

Salesforce’s vision for the next generation of Sales Cloud is much more active.

The company described a CRM that can capture information, update itself, identify what should happen next, and increasingly execute work on behalf of the seller.

That moves CRM from being primarily a system of record toward becoming more of a system of action.

Salesforce described an outcome-based sales agent that can work across the deal cycle, helping sellers with activities from prospecting through closing and account growth.

That doesn’t mean the salesperson disappears.

It means more of the administrative and analytical work surrounding the salesperson can potentially happen without requiring the rep to manually initiate every step.

For sales leaders, that could mean less time spent asking whether Salesforce has been updated and more time focused on whether the right actions are actually happening.

Claude Becomes Part of the Sales Team

Day 2 also gave Salesforce more opportunity to demonstrate what its expanded relationship with Anthropic could mean for sellers.

With Salesforce in Claude, employees can interact with Salesforce data and processes directly through Claude rather than constantly moving between an AI assistant and CRM.

That distinction matters.

The value isn’t simply asking Claude to draft an email or summarize a document.

The larger opportunity comes when AI has access to governed business context.

Imagine preparing for an account meeting and asking Claude to review the account, opportunity history, recent activity, open issues, and relevant customer context before recommending what deserves attention.

Or asking it to help identify stalled opportunities and then take an approved action back in Salesforce.

That is very different from using a standalone generative AI tool with information manually copied into a prompt.

Salesforce’s broader AIforce strategy is designed to make Salesforce’s data, workflows, business logic, permissions, and actions available to AI interfaces like Claude while keeping Salesforce as the governed business foundation underneath them.

Agentforce Has Entered Its ROI Phase

The dedicated Agentforce keynote made perhaps the clearest change in Salesforce’s messaging.

The central question was no longer:

What can an AI agent do?

It was:

Can an AI agent create measurable business value?

Salesforce framed the keynote around moving organizations from a fast start to real ROI and said approximately 30,000 customers are now live on the Agentforce platform.

Rather than relying primarily on futuristic demos, Salesforce spent significant time highlighting customer use cases and measurable operational outcomes.

That shift is important.

Enterprise AI projects ultimately have to compete for the same budgets as every other technology initiative. A compelling demonstration isn’t enough. Companies need to understand whether an agent can reduce cost, increase productivity, improve customer experience, generate revenue, or remove enough manual work to justify the investment.

Salesforce introduced an Agentic ROI Playbook centered around getting agents into production, monitoring their performance, and continuously improving them.

For businesses considering Agentforce, that creates a much better starting question than “Where can we use AI?”

Instead, ask:

Which business outcome are we trying to improve, and how will we know whether the agent improved it?

Agentforce Is Becoming More Than a Short Conversation

Another interesting theme was the evolution toward agents capable of working toward longer-running goals.

Salesforce demonstrated a model that incorporates goals, planning, signals, memory, durable execution, and continuous refinement.

But importantly, Salesforce continued to emphasize a human in the loop.

That distinction matters.

The future Salesforce demonstrated isn’t necessarily one where an agent receives unlimited autonomy and disappears for hours to make business decisions on its own.

Instead, agents can potentially work across longer processes while humans retain opportunities to review, approve, redirect, or intervene.

For Revenue Operations teams, that introduces an entirely new process-design question.

Today we frequently ask:

What should be automated?

Increasingly, we’ll also need to ask:

What should AI be allowed to decide, what should it recommend, and what still requires human approval?

Those are not purely technology decisions.

They’re governance decisions.

Managing Agents After Deployment Becomes a New Discipline

One of the more practical demonstrations from the Agentforce keynote involved Agent Optimizer.

Building an agent is only the beginning.

Once an agent is operating in production, organizations need to understand whether it is producing the expected outcomes and where it is failing.

In Salesforce’s example, Agent Optimizer identified that refund requests above a particular threshold were being escalated at an unusually high rate.

Rather than simply reporting the problem, the system traced it back to the agent configuration, recommended changes, and helped generate tests for the proposed fix.

This is an important evolution.

As companies deploy more agents, somebody will have to own their ongoing performance.

Organizations already monitor sales processes, integrations, automations, data quality, and system adoption.

Agent performance may soon become another operational responsibility.

For RevOps teams, that could mean measuring things like:

• Agent completion rates
• Escalation rates
• Human intervention
• Accuracy
• Cost per interaction
• Time saved
• Revenue influenced
• Customer outcomes

The AI strategy doesn’t end when the agent goes live.

That may actually be when the real work begins.

Revenue Cloud Is Becoming an Agentic Revenue Platform

Revenue Cloud also received significant attention on Day 2.

Salesforce positioned Revenue Cloud as a unified revenue engine spanning pricing, quoting, contracting, and billing.

This continues a transition we’ve been watching closely as Salesforce moves beyond the traditional Salesforce CPQ model toward a broader product-to-cash platform.

But AI adds another dimension.

Instead of simply automating steps within quoting and billing, Salesforce is increasingly talking about agents participating in the revenue process itself.

That could include helping sellers configure solutions, guiding pricing decisions, identifying relevant products, accelerating quotes, supporting contracting, or helping manage more complex revenue models.

For organizations currently evaluating the future of Salesforce CPQ, this makes the decision bigger than simply determining how to migrate existing functionality.

The better question may be:

If we were designing our product-to-cash process today, knowing what these platforms can now do, would we design it the same way?

In many organizations, the answer will probably be no.

Data 360: Context Becomes the Competitive Advantage

If agents are going to make more decisions and take more actions, they need more than a language model.

They need context.

That was the central message behind the Day 2 Data 360 conversation.

Salesforce’s argument is straightforward: many companies may eventually have access to similar AI models. The differentiator becomes the proprietary business context those models can safely access.

That context might include:

Customer records.

Opportunities.

Support cases.

Orders.

Invoices.

Marketing engagement.

Product usage.

Contracts.

Emails.

Call transcripts.

Knowledge articles.

Slack conversations.

And information from systems outside Salesforce.

Data 360 is intended to help unify and govern that information so humans and agents can retrieve the right context when they need it.

Importantly, that doesn’t mean simply dumping every piece of company data into an AI prompt.

Salesforce’s Data 360 architecture focuses on identifying the relevant and authorized information for a particular interaction while accounting for factors like permissions, freshness, privacy, residency, and cost.

For organizations racing toward AI, this reinforces something we have discussed repeatedly:

Your AI strategy is also a data strategy.

Real Agentforce Deployments Are Starting to Matter More Than Demos

Day 2 also included more examples of companies putting agents into production.

Live Nation, for example, announced an expansion of its Agentforce deployment following a pilot at BottleRock Napa Valley.

Its Melody agent helped festival attendees answer questions about schedules, stages, food, and other event information. Salesforce says the pilot generated more than 37,000 interactions and 17,000 customer service sessions during the launch window, with 85% of users getting the information they needed within three responses.

Live Nation is now expanding the concept with a Venue Agent designed to provide 24/7 information across U.S. concert venues.

The interesting part isn’t simply that Live Nation built a chatbot.

It’s how the architecture works.

Rather than creating a completely separate agent for every venue, the system uses Salesforce and Data Cloud to determine which information should be used for the specific venue and interaction. When the agent can’t appropriately handle a request, the conversation can be routed to a human team through Slack.

That combination of shared architecture, contextual data, governance, and human escalation is much closer to what enterprise AI will probably look like in practice.

AIforce Ties the Pieces Together

Day 1 introduced AIforce as Salesforce’s new live interface layer.

Day 2 provided more context for how all of these pieces fit together.

Salesforce increasingly appears to be separating the business platform from the interface used to access it.

The foundation still includes Salesforce data, metadata, permissions, workflows, business logic, security, and governance.

But the interface might be:

Salesforce.

Claude.

Slack.

Agentforce Coworker.

Or another AI experience.

That could become one of the most significant changes to Salesforce architecture in years.

For decades, Salesforce implementations have involved considerable discussion about page layouts, navigation, record pages, tabs, components, and user experience.

Those things aren’t disappearing tomorrow.

But if an employee can simply describe what they need and have an intelligent interface assembled around that task, the role of the traditional CRM user interface begins to change.

Day 2 Reinforced the Importance of the Foundation

There is an interesting contradiction running through Dreamforce 2026.

AI makes it look easier than ever to build sophisticated experiences.

But making those experiences reliable may require organizations to become more disciplined underneath them.

An agent needs good data.

It needs clear processes.

It needs permissions.

It needs defined business rules.

It needs governance.

It needs measurable objectives.

And it needs clearly defined boundaries around what it can and cannot do.

If your organization doesn’t know when an opportunity should advance today, adding an AI agent won’t magically create a good qualification methodology.

If your account data is unreliable, giving more AI tools access to it doesn’t make the data reliable.

If ownership between Sales, Service, Marketing, and Finance is unclear, agents may simply automate that ambiguity faster.

That may be one of the most important lessons coming out of Dreamforce this year.

What Revenue Leaders Should Take Away From Day 2

Day 1 showed us where Salesforce believes enterprise software is going.

Day 2 was about making that vision operational.

For Revenue Operations leaders, there are several questions worth bringing back to the business:

Where are people spending time on work that doesn’t require human judgment?

Those activities may be strong candidates for automation or agents.

Where would better customer context materially improve a decision?

Those may be opportunities for Data 360 and AI-powered reasoning.

Which decisions should AI recommend versus execute?

That distinction should be defined before agents are deployed.

How will we measure whether an agent is successful?

Every meaningful AI initiative should have an expected business outcome.

Is the underlying revenue process actually ready?

This may be the most important question of all.

The Bigger Dreamforce 2026 Story Is Taking Shape

After two days, the broader Dreamforce message is becoming clearer.

Day 1 was about access: bringing Salesforce intelligence into Claude, Slack, Lightning, and other interfaces.

Day 2 was about execution: allowing agents to participate in sales, service, revenue management, and other business processes while demonstrating measurable results.

Underneath both is the same foundation: trusted customer data, business context, permissions, governance, and clearly defined processes.

The interface may be changing.

The importance of getting the underlying revenue operation right is not.

In fact, as AI takes on more responsibility, it becomes even more important.

For companies evaluating Agentforce, Data 360, Revenue Cloud, Claudeforce, or the broader Salesforce AI ecosystem, the first step shouldn’t necessarily be selecting the newest technology.

It should be determining where AI can create measurable value, whether the underlying process is ready, and whether the data and governance foundation can support it.

That’s the work that turns an impressive Dreamforce demo into something that can actually operate inside a business.

Ready to evaluate where AI fits into your Salesforce and revenue strategy? Revenue Ops can help you assess your processes, data, architecture, and AI readiness before you make the next investment.