At Dreamforce 2026, the main keynote spent almost no time inside Sales Cloud or Service Cloud. Instead, Salesforce launched AIforce, which brings Salesforce data and business logic into AI tools like Claude and Slack, governed by the permissions your company has already set. People don’t have to go into Salesforce to get work done anymore. Salesforce comes to them.
That matters in manufacturing, where much of your revenue and service runs through dealers, distributors, field reps, and customers who rarely open Salesforce. In the keynotes I sat in on, the manufacturing announcements kept putting an agent in front of those people.
When a dealer texts an agent to ask how close they are to the next incentive tier, the answer goes straight to the dealer, with nobody at your company checking it first. The keynote demos made this look easy. In one, a sales leader worked entirely in a screen Claude built from Salesforce data.
But you can’t just go to Claude and say, “Be a CRM.” If the data isn’t organized, correct, and connected, the AI doesn’t work. Everything you already know about hierarchies, rebates, and forecasting still matters. How you put it to work is what’s different.
What Salesforce showed manufacturers
Dealers and distributors get answers without logging in
Salesforce showed dealer agents that answer on partner sites or over SMS and WhatsApp, with no login. A dealer can ask where they stand on a rebate program or how much more to order to reach the next tier. Those answers come from rebate and purchase data that often lives in the ERP, so connecting it, through Salesforce’s data layer or a platform like Snowflake, belongs in the pilot plan from day one. Before go-live, decide how the agent confirms who’s texting and what each dealer can see, so one distributor never sees another’s pricing.
Field reps work from the truck
Reps can ask Salesforce’s sales concierge about open opportunities from Slack, Teams, or WhatsApp, get a spoken briefing in the mobile app before a visit, and dictate notes afterward. Reps working in Claude also get prebuilt sales skills that build account plans and close plans. They’re built to Salesforce’s spec, so if your process differs, edit them to match. A close plan built on stale data shows reps the wrong number, and once they see it, they stop trusting the tool and go back to their manual ways.
Customers and partners get help after the sale
Manufacturing Cloud’s AI Warranty Concierge guides customers through a claim, and Casey, a prebuilt Salesforce service agent, resolves issues for customers, partners, and distributors. You pay for Casey only when it resolves an issue without a handoff. Salesforce’s new Agent Health Monitoring flags wrong answers that look right so your team can fix the agent fast. Decide in advance who the agent hands off to, and who makes it right when a customer or dealer acts on a wrong answer.
Back-office processes get an agent of their own
Marshall is the agent behind Agentforce Operations. You describe a process like distributor onboarding, and Marshall builds and runs it across your systems, including customized SAP and Oracle. It automates what’s possible and leaves for a person what a person needs to do.
Getting started begins with a focused process analysis to see where Marshall fits. Because Marshall is designed to scale, start with a small, easily monitored workflow, such as customer onboarding or product return requests.
What to plan for in 2027
In my Dreamforce meetings, nobody asked to turn any AI tools on right away. People wanted to know what’s possible, and they had no idea what it would cost. In one conversation, a manufacturing leader loved the idea of having an agent in Claude build forecast graphs on demand — until they realized they had no idea how or where those consumption charges would be billed.
Here’s what I’d tell you:
- Budget for data work and AI consumption: Pricing for most new agents isn’t published, so build in room, and fund cleanup of the hierarchies and rebate data your agents will read.
- Pilot before you scale: For some agents, the time from idea to install is days, not months, so a small test is inexpensive. Pick a few use cases tied to a number you already track, like dealer call volume or warranty claim cycle time.
- Start with yourself, then your people: Use AI tools yourself first, then map what each role does on a typical day. Walk through a process manually to see where tools like Claude or Agentforce can handle specific steps.
A golfer carries up to 14 clubs and doesn’t swing every one on every hole. In the bunker, they reach for the sand wedge, not the putter. Salesforce just filled your bag with new agents and tools. Pick the ones that fit each role’s situation, and don’t feel you have to use them all. Then get the data behind them right, so when a dealer texts about the next incentive tier, your company can stand behind the answer.
Catch up on the rest of Dreamforce 2026 in our full recap. If you’re sorting out which agents fit which roles in your operation, book an AI Pathfinder, a free strategy session. Bring the use cases your team is weighing, and you’ll get back a recommended order for where to start.
