Many organizations have moved past experimenting with AI. The early demos are done, the possibilities are clear, and the pressure to deliver something real is building.
That pressure shows up in practical ways: service teams want agents that can answer from current knowledge articles, sales teams want account-specific summaries and follow-up drafts, and digital teams want conversational experiences that can point customers to the right information without forcing them to search across disconnected systems. The harder part begins when those early ideas need to work inside actual systems, with real data, for real users.
At the same time, expectations have shifted. People are used to asking questions in plain language and getting direct answers. In enterprise environments, that expectation runs into a different reality: fragmented knowledge, access controls, and systems that weren’t designed to work together.
Salesforce Agentforce has become a focal point for many of these efforts. It gives teams a way to build agents that can take action across workflows. But once teams start working with it, a familiar question comes up: what should the agent actually do, and how do you make sure it produces answers people can rely on?
Where Complexity Shows Up: Fragmented Content, Custom Workflows, and Trust
When Coveo began working with Agentforce, the goal was to support a wide range of use cases. Agents could answer questions, generate content, assist with workflows, or hand context to a human agent. That flexibility quickly surfaced a pattern: every organization wanted to use the same core capability in a slightly different way.
“What we found over that last year is that by doing POCs, every customer wants their own flavor,” said Mathieu Lavoie-Sabourin, Product Manager, Coveo. “They want to customize it their own way, they want to build their own flow, their own approach.”
That’s what makes Agentforce appealing. It also makes it harder to standardize how these projects get off the ground.
Each team brings its own data, its own workflows, and its own expectations. Without a consistent way to connect those pieces—and without a clear path from an initial prototype to something that can be sustained—projects can slow down.
That is where the Slalom partnership becomes important. Coveo can provide the retrieval foundation agents need, while Slalom helps customers shape that foundation into Salesforce workflows that match their business, data, and users.
Why Partners Matter Here
For many organizations, the challenge isn’t deciding whether to invest in agentic AI. It’s having the mix of skills needed to implement it effectively.
These projects cut across multiple areas: Salesforce architecture, data access, retrieval systems, and user experience. Most teams are still building those capabilities internally.
Working with partners brings in that experience earlier. It also shortens the path to something usable.
Slalom and Coveo are approaching this from two sides. Slalom focuses on how Agentforce gets implemented inside Salesforce environments. Coveo provides the retrieval layer that connects agents to the content they need to operate in context.
The result is less about adding another layer and more about making the first steps easier to take—and easier to build on.
A Structured Way to Get Started
The joint approach centers on a short, focused engagement.
Slalom’s Agentforce Launchpad begins with a three-week proof of concept in a Salesforce sandbox. The scope is intentionally narrow: one use case, one agent, and a defined set of data.
That agent might support a service scenario, drawing from knowledge articles, or a sales scenario, using account and opportunity data. The version developed with Coveo adds a specific constraint: the agent is grounded in content that has already been indexed.
“One sales or service agent configured within a Salesforce sandbox, grounded in one public content source indexed via Coveo,” said Eva Klug, Principal Consultant at Slalom.
That setup keeps the initial work focused while still producing something that can be tested with real inputs. It also avoids the need to rebuild or relocate existing knowledge before getting started. For existing Coveo customers, that can mean using content they have already indexed. For organizations new to Coveo, it provides a faster way to connect Agentforce to approved sources without turning the proof of concept into a content migration project.
“We’re meeting those customers where they’re at,” Klug said, “and taking advantage of what they’ve already invested in in Coveo.”
From there, the implementation can expand—adding more sources, introducing additional workflows, or moving toward production.
Grounding the Experience in Real Enterprise Content
One of the practical challenges in building AI-driven experiences is deciding what information the system should rely on.
In most organizations, knowledge is spread across systems—public websites, documentation, internal knowledge bases, and more. Pulling that together into something usable by an agent is not always straightforward.
The approach here is to leave that content where it is and make it accessible at the moment it’s needed.
When a user asks a question, the agent retrieves relevant passages from indexed sources and uses them to generate a response. That retrieval step shapes what the system can say.
This becomes more important as usage grows. Inaccurate responses and hallucinations are related, but not identical: an answer can be wrong because it uses stale or incomplete context, while a hallucination is when the system generates information that is not grounded in the available source material. Both problems become more likely when the agent cannot retrieve the right enterprise content at the right moment.
By grounding responses in known content, the system stays closer to what the business already understands to be true.
What This Looks Like in Practice
During the session, Slalom walked through a working example.
In the demo, a user begins by asking a general question through a chatbot interface. The agent retrieves relevant content, indexed through Coveo, and uses it to generate a response. The sources are visible alongside the answer, which helps show not only what the answer is, but where it came from.
The user then refines the question and asks for more specific information. The agent responds again, drawing from the same content base while adjusting the answer based on the updated context.
At that point, the interaction shifts. The user decides to speak with a person. The conversation is routed to a human agent inside Salesforce, along with the full history of what was discussed.
From the agent’s perspective, the same content layer is still in play. Suggested responses are generated using those indexed sources, helping the agent respond more quickly without starting from scratch.
The example then continues into a sales workflow. The interaction becomes a lead, then an opportunity. A sales user opens that opportunity and asks the system to summarize the offering for that specific account.
The response combines two inputs: retrieved content and Salesforce data tied to the account. The result is a summary tailored to that context.
From there, the system drafts a follow-up email based on that summary.
The flow moves from customer interaction to internal support to sales follow-up, all using the same underlying content.
From Initial Use Case to Broader Adoption
The three-week engagement is meant to produce something concrete, but it’s not the endpoint.
Once teams see how the agent behaves with real inputs, they often identify additional use cases—more content to include, more workflows to support, or ways to connect the system more closely to production processes.
Each step builds on the previous one. The initial structure makes it easier to extend the system without starting over.
From a delivery perspective, this is where both the implementation model and the underlying content layer matter. The same factors that make early prototypes flexible can make them difficult to expand without a consistent foundation.
Bringing It Together
Agentforce gives organizations a way to build agents inside Salesforce. The challenge is turning that capability into something that fits real systems and workflows.
Slalom and Coveo approach that problem from different angles. One focuses on how the system is built and deployed. The other connects it to the information it needs to operate in context.
Together, they create a way to move from an initial use case to something that can be used more broadly—without needing to rework the foundation each time.
For teams working through their first Agentforce projects, that combination can help turn an early concept into something people actually use.