Salesforce’s new AI agent shows what becomes possible when ERP and CRM data lives in the same place. For ERP buyers, that has implications worth thinking through. 

AI agents are only as useful as the data they can reach. The real question isn’t whether AI is here. It’s whether it has access to the data needed to surface meaningful insight.

Salesforce’s Agentforce Coworker is the most pointed example yet of why that question matters. It’s an AI assistant designed to act as a coworker inside the tools your team already uses, reasoning across whatever data lives inside the Salesforce Platform. For manufacturers whose ERP is built natively on Salesforce, that “whatever data lives inside” includes production records, inventory, work orders, purchase orders, and sales orders alongside the CRM. For those whose ERP runs on a separate platform connected to Salesforce via middleware, it doesn’t. 

This blog walks through what Agentforce Coworker is, what becomes possible when it can reach ERP data natively, and why this is the moment manufacturers should be reevaluating where their ERP is built. 

agentforce_coworker_watch_demo_banner_v2

What is Agentforce Coworker? 

Agentforce Coworker is Salesforce’s AI agent, positioned as a teammate that operates inside the Salesforce environment your team already uses. It’s part of the broader Agentforce platform and is built for the kind of work where a real person would normally open multiple systems, correlate information by hand, and form a recommendation. 

Where most AI assistants are limited to a single application or a single domain of data, Agentforce Coworker can reach across any application in the Salesforce Platform: sales records, service tickets, marketing data, and, if your ERP is built natively on Salesforce, your production, inventory, and procurement data alongside it. 

Salesforce CEO Marc Benioff has called it a defining moment in Salesforce’s broader AI strategy.

The architecture that makes this work 

This is the part most ERP buyers don’t fully understand until they need it. 

Agentforce Coworker can reach data that lives inside the Salesforce Platform. It does not natively reach data sitting in a separate ERP database connected to Salesforce via middleware, third party connectors, or scheduled syncs. Each of those approaches introduces latency, integration cost, and gaps between what the CRM knows and what the ERP knows. 

Rootstock ERP is built natively on the Salesforce Platform. Production data, inventory records, work orders, BOMs, financial data, and procurement data all live in the same Salesforce data model as your CRM records. When Agentforce Coworker is asked a question that crosses the boundry between CRM and ERP, it doesn’t have to call an external API, wait for a sync, or translate between systems. It reads both as if they were the same dataset, because architecturally, they are. 

What’s striking is that Agentforce Coworker doesn’t need to be pre-configured with knowledge of Rootstock objects. It explores the Salesforce metadata available to it, automatically identifies the right ERP objects to query, and works from there. The same way it operates on any native Salesforce object. 

For manufacturers, the practical consequence is this: your ERP architecture determines whether Salesforce’s ongoing AI investments compound for you or pass you by. Salesforce invests billions of dollars annually in this space. Each new agent capability, model improvement, and platform integration reaches platform native ERP data the moment it ships. For ERPs that live elsewhere and connect via middleware, every new Salesforce AI capability requires custom integration work to be useful, and the gap between the two paths widens with every release. 

How this looks in practice 

Andrew Russo, VP of Business Systems at BACA Systems, a Rootstock customer, ran Agentforce Coworker against his Rootstock ERP data and shared the experience publicly. The demo has been circulated widely, including by Salesforce CEO Marc Benioff. 

In Andrew’s example, he opens a conversation by asking Agentforce Coworker which spare parts his customers tend to buy together. What follows is not one answer but a working dialogue. Agentforce Coworker independently discovers Rootstock’s data model, performs a real market basket analysis, identifies a set of customers who bought part of a common bundle but not all of it (a clear upsell pattern), then cross checks the campaign against current inventory and proactively warns Andrew that the constrained part will run out if the campaign succeeds. It queries open purchase orders, confirms nothing is in the pipeline, surfaces the historical reorder cadence, and recommends placing a reorder before outreach begins. 

In a single conversation, Agentforce Coworker moves across sales analysis, CRM customer identification, inventory verification, supply chain risk forecasting, and procurement planning. That’s the kind of cross functional question that traditionally required swivel chairing between five systems, an analyst, and a stack of spreadsheets. 

Andrew’s takeaway, in his own words: 

“I’ve been trying to break Agentforce Coworker and ask questions it can’t figure out the answer to and I’m actually struggling. Every time I think I have found a way to stump it, Coworker figures it out and it responds with the answer.” 

You can watch Andrew’s original post here and Benioff’s repost here. 

What other manufacturing questions become possible 

Andrew’s spare parts example is one instance of a much broader category. Any question that historically required correlating data from CRM, ERP, and back-office systems becomes a single conversation when those systems share a data model. A few examples worth considering: 

  • Customer service: What’s the production status of this customer’s order? Are we on track to ship, and if not, what’s the new ETA? 
  • Sales operations: Which open opportunities in my pipeline depend on parts that are below safety stock? Show me the deal value at risk. 
  • Account management: Pull every active sales order, RMA, support ticket, and unpaid invoice for this customer and tell me where the relationship stands. 
  • Operations leadership: What are the top five constrained parts right now, and which customer orders and quotes depend on them? 
  • Finance: How is gross margin trending on our configured products? Where is margin eroding, and on which product lines? 
  • Supplier management: Which active suppliers are running behind on commitments? Which production runs depend on them? 
  • Forecasting: Compare actual demand for our top spare parts against forecast. Where are we systematically missing? 

These are just a starting point. The most valuable questions are the ones only you and your team can think of, the ones specific to how your business actually runs. 

Operations, sales operations, finance, and customer-facing teams ask questions like these every day. With an AI agent that can read both CRM and ERP data, the same question returns the answer, the context, and the next recommended action. 

What this means for ERP buyers 

Manufacturers evaluating ERP today are making a decision that will shape their AI capability for years, not just the year of go live. The platforms that compound AI value over time are the ones where data lives in the same place as the AI model. Platforms that require middleware to connect their ERP to AI will get more capable in narrow ways, through purpose-built integrations and dedicated AI features, but will not benefit from broad platform AI investments the way platform native ERPs do. 

The Agentforce Coworker moment isn’t really about a single AI agent. It’s about what becomes possible when the underlying platform is architected to make every future AI agent immediately useful. For Rootstock customers, that’s already true. For others, it’s the question to be asking now. 

Continue the conversation 

To see Agentforce Coworker running on live Rootstock ERP data, watch Andrew Russo’s post. Rootstock also shared a LinkedIn breakdown of what Andrew demonstrated. 

If you’re thinking about how to position your ERP for the AI era, this is a useful conversation to be in.