Dreamforce 2026 Is Over. Here Are the 5 Decisions Your Leadership Team Now Has to Make About AI Agents

Dreamforce 2026 Is Over. Here Are the 5 Decisions Your Leadership Team Now Has to Make About AI Agents

Dreamforce 2026 wrapped up with one message for senior leaders: the experiment phase of enterprise AI is ending. Salesforce framed the event as taking production agents to the next level, with the long-term-goal agents and the debut of its own reasoning model as headline items. 

If you run a business, you don't need a feature list. You need to know what is worth acting on now, what to wait on, and what could go wrong. This post gives you that.

What actually changed

1. AIforce: Salesforce goes where your people already work.
AIforce brings Salesforce data, workflows and business context to the surface you prefer, whether that is Claude, Slack or Salesforce itself, under the permissions you have already defined. It includes Claudeforce inside Claude, Slackforce inside Slack and Agentforce Coworker inside Salesforce. For a CEO this means your CRM is no longer a system your team has to log into. It becomes a layer that works across the tools they already use.

2. Agentforce Coworker: an AI teammate, available now.
Coworker is an autonomous teammate built into Salesforce. It answers natural-language questions across your CRM, Slack and other connected sources, and when asked to get something done it builds a plan and executes it, orchestrating other agents. It is generally available today. The early adopter numbers are worth noting: Fulton Bank went from zero to more than 20 use cases and around 3,000 users within weeks, and Adecco is rolling Coworker out to 27,000 employees across more than 40 countries. 

3. Long-horizon agents and Koa: AI that pursues goals, not tasks.
Long-horizon agents pursue a goal over days, weeks or months. You might assign "re-engage my at-risk deals" and they build a multi-step plan, checking in for approval based on guardrails you define. Koa is Salesforce's first CRM reasoning model, built on NVIDIA Nemotron. It is trained entirely on synthetic scenarios, so no real customer data trains it. Be aware that timing matters here: long-horizon agents are in pilot with general availability planned for November 2026, and Koa is in pilot.

What did not change

One point matters for your budget. Salesforce announced no new Agentforce pricing at Dreamforce. That means your cost conversation depends on your current contract, your data volume and how many agents you deploy. Don't assume the headline features are included in your current licence. Ask your account team or partner for a written answer. 

The 5 decisions for your next leadership meeting

Decision 1: Which one business outcome do we want agents to own?
Don't start with "where can we use AI?" Start with one measurable outcome, such as faster lead response, lower cost per service case or shorter quote-to-cash. Siemens, for example, uses Agentforce to qualify inbound leads for its 18,000 sellers. That is one narrow, measurable job done at scale. Pick yours. 

Decision 2: Is our data ready to be trusted by an agent?
An agent acts on whatever your CRM holds. Duplicate accounts, stale contacts and inconsistent fields that a human works around will cause an agent to act wrongly, at speed. Before you buy anything, ask your sales and operations heads one question: if an agent acted on our CRM data today, would we be comfortable with the result?

Decision 3: Who is accountable when an agent gets it wrong?
This is a governance question, not an IT one. Name an executive owner for agent performance. Define what an agent may do alone and what needs human approval. Salesforce is building tools for this: Agent Health Monitoring is generally available with 16+ built-in metrics, and Agent Optimizer, generally available in October, finds recurring failure patterns in production sessions and ranks them by impact. Tools help, but a named owner matters more. 

Decision 4: Build, buy or wait?
You have three realistic options.

• Act now: use what is generally available, such as Agentforce Coworker and most job-ready agents, on a narrow use case.
• Pilot: join pilots for long-horizon agents or Koa if the use case is strategic and you accept the risk of early software.
• Wait: hold off on anything still in beta, such as Claudeforce, until your data and governance are in place.

Note that Hunter, the outbound sales agent, is not generally available until November. Don't plan around dates that haven't arrived. 

Decision 5: How will we measure return after 90 days?
Set the scorecard before you launch. Pick two or three numbers, such as response time, cases resolved without escalation or pipeline created, and compare them to a baseline you record today. Without a baseline you will not be able to defend the investment to your board.

Risks to raise in the room

• Pilot sprawl. Many teams launching agents independently creates cost and security exposure. Centralise ownership.
• Over-trust. Salesforce itself describes the scariest agent failures as the quiet ones, where the agent looks fine but the answer is wrong. Build review into the process.
• Vendor concentration. More AI inside one platform increases convenience and dependence together. Keep your data portable.
• Third-party connections. Salesforce introduced MCP Risk Scores to assess servers registered with Agentforce for risks such as prompt injection and tool poisoning. Make sure someone is reviewing what your agents connect to.

Dreamforce 2026 did not ask leaders to bet on AI. It showed that the platform is ready for production, and that the competitive gap will now come from execution: clean data, clear ownership and focused use cases. Companies that decide this quarter will have a real result to show by the new year. Those that wait will be buying the same tools later, with less experience.

Frequently Asked Questions

What is AIforce, and why should a CEO care?

AIforce is Salesforce's new layer that brings your Salesforce data, workflows and business context to the surface your team prefers, whether that is Claude, Slack or Salesforce itself, under the permissions you have already defined. For leadership, it means the CRM stops being a system people must log into and becomes something that works inside the tools they already use.

Did Salesforce announce new Agentforce pricing?

No new Agentforce pricing was announced at Dreamforce 2026. Your cost depends on your current contract, your data volume and the number of agents you deploy, so get a written answer from your account team or implementation partner before you commit.

Should we act now or wait?

It depends on the use case. Tools that are generally available, such as Agentforce Coworker, suit a narrow, measurable workflow today. Pilot features, such as long-horizon agents and Koa, suit strategic use cases where you accept early-software risk. Anything in beta can wait until your data and governance are ready.

Does our CRM data need to be clean before we use AI agents?

Yes, this is the most common blocker. An agent acts on whatever your CRM holds, so duplicate accounts, stale contacts and inconsistent fields lead to wrong actions at speed. A data audit should come before any agent launch.

Who in the company should own AI agents?

A named executive, not just IT. This person is accountable for agent performance, decides what an agent may do alone and what needs human approval, and reports results to the leadership team.