AI Agents in Manufacturing

At Dreamforce 2026 in San Francisco (September 15 to 17), Salesforce introduced AIforce, among other announcements. The idea behind it: Salesforce data, business logic, workflows, and governance should be available not only within Salesforce, but also in other work environments such as Slack, Claude, or Salesforce Lightning. At launch, Salesforce presented three approaches to make this possible: Claudeforce, Slackforce, and Agentforce Coworker.

For manufacturing companies, this means a fundamentally new way of accessing AI:

• Work no longer happens exclusively in the CRM.

• Data, permissions, business rules, and workflows can increasingly be used through other interfaces and AI agents.

• Salesforce Lightning remains, but it is now one of many ways to access these capabilities.

What matters is that the processes behind them are sound. Which information is needed? Which rules apply? And when does a person need to decide or approve something? Once these questions are answered, the work environment can be better adapted to everyday work. A service or sales team could, for example, see customer information and the next approved steps right where it coordinates, without constantly switching between applications.

This saves steps and can simplify workflows. However, ERP, machine, or engineering data is not automatically connected as a result. That still requires suitable interfaces, controlled access rights, and clearly defined processes. Yet the real value goes beyond simply accessing information. AI can not only help find answers, but also take on concrete tasks within a work process.

Generative AI can answer questions, formulate texts, and create drafts. AI agents go one step further: within clear rules, they can take on tasks within a process themselves.

In manufacturing, an agent could, for example, check whether a request relates to a particular machine and an existing contract, gather the relevant service information, and prepare a spare-parts order. If the cost exceeds a defined threshold or important information is missing, the process pauses and asks an employee for approval.

This does not mean that companies hand their decisions over to AI. Rather, the work shifts: routine tasks and information searches decrease, while technical assessment, accountability to customers, and the handling of exceptions remain with people.

At Dreamforce, Salesforce also announced preconfigured agents for service, sales, and supply chain. Which features are actually available depends on the specific product, the region, and the contract.

But what does this look like in practice? In manufacturing, many processes require information from several sources to come together before someone can continue working. Three examples show where AI agents can help, from the first review of a tender to product discovery and technical service.

1. Assessing RFPs Faster

   When responding to extensive requests for proposal (RFPs), sales, engineering, and the bid team often have to decide under time pressure: Can we meet the requirements, and is it worth submitting a proposal?

   
   

   Here, AI can handle the initial review. It structures the requirements, compares them with existing information, and prepares a first assessment. The decision on whether the company submits a bid remains with the responsible employees. The technical review by engineers also remains essential. AI helps them mainly by quickly identifying the points that need closer examination.

   
   

   The benefit: the bid team can use its time more deliberately, and scarce technical expertise is brought into the critical questions earlier.

2. Guiding Customers from Application to the Right Product

   Many industrial websites are designed on the assumption that customers already know a product and search for it specifically. With complex product ranges, that is often not the case. Customers know which problem they want to solve, but not necessarily the right product or the right technical filters. Find out how ifm approaches this challenge with an AI Product Advisor here.

   
   

   Traditional search functions quickly reach their limits here. Language models, by contrast, can first capture the specific need and narrow down, through targeted follow-up questions, which solution is suitable. A pure product search thus becomes advisory guidance that starts with the application.

   
   

   The benefit: customers find a suitable solution more easily. At the same time, a general question is more likely to turn into a specific, qualified inquiry.

3. Making Service Knowledge Available Faster

   In technical service, the right details matter:

• Which machine type is affected?

• Which serial number belongs to it?

• What does the contract cover?

• Which service cases have occurred so far?

• Which technical documentation is relevant?

  
  

  When this information comes together in the customer process, service and sales teams can give well-founded answers more quickly, for example about the right spare-part component or a technically suitable upgrade.

  
  

  A concrete example is the expanded partnership between Salesforce and Siemens. The plan is to connect Siemens Teamcenter Service Lifecycle Management with Agentforce, to make technical product information available across sales, service, and customer processes. Salesforce also describes the use of Agentforce to process incoming leads at Siemens.

These examples shows how such approaches can look at a large industrial company. Whether these solutions can be transferred to other manufacturing companies without significant adaptation is, however, not yet known.

An AI agent can only work as reliably as the information, rules, and access rights it can draw on allow. If machine, contract, and service data are not cleanly linked, the results quickly become incomplete or contradictory. And if it is not possible to trace what the agent did and why, trust in the solution suffers.

That is why the starting point does not need to be a company-wide AI project. It makes more sense to begin with a clearly defined process:

1. Choose a specific use case. For example, preparing a spare-parts order or conducting an initial review of an RFP.

2. Bring the required information together. Only the CRM, ERP, machine, and knowledge data that this process actually requires.

3. Set clear boundaries. Which tasks may the agent prepare? Which may it execute itself? And when must a person decide or approve?

4. Make the value measurable. Use suitable metrics such as time to a qualified response, first-time fix rate, or end-to-end process cycle time.

5. Review results and scale gradually. Only when outcomes, error rates, and exceptions hold up in day-to-day operations should additional processes be added.

Dreamforce 2026 points in a clear direction: Salesforce wants data, business logic, and AI agents to be available where employees do their work, not only within the CRM.

The real value, however, does not come from introducing an AI agent. It comes when previously scattered information is brought together, processes become simpler, and employees reach reliable answers faster. At the same time, it must remain clear who makes which decisions and who bears responsibility.

The key question is therefore not: "Where can we use AI?" Rather: "Where are our customers or teams losing time today because information has to be searched for, entered more than once, or passed between departments?"

Companies that start exactly there, choose a manageable process, and check success using concrete metrics build a solid foundation for further steps. The goal is not as much AI as possible at once, but a noticeable improvement in everyday work.