Content teams are entering a new phase of responsibility, yet many organizations haven’t recognized this shift. At Salesforce, the scope and impact of content are expanding because of AI, rather than diminishing. We are no longer just publishers of documentation; we are the architects and stewards of the data that makes AI functional.

Our AI agent is having millions of conversations with customers on Salesforce Help, and right now, those are based predominantly on content delivered by our Content Experience (CX) team. The customer may never read the original article on which an AI conversation is based, but the content still grounds and informs the answer. As Jo Ward, Salesforce SVP of CX, puts it, “We’re building the foundational content that powers our AI-driven initiatives.”

This reality raises a practical question: How should content teams evolve when their output is increasingly consumed by both AI systems and humans? To answer this, we have to look beyond tools and investigate how the very nature of our work is shifting. 

Studying the transformation from the inside

At Salesforce, we approached the question the way we approach any research problem: by studying it before trying to solve it. Salesforce is unusual in that our CX organization includes an embedded researcher. Instead of observing transformation from the outside or jumping straight to tool deployment, we set out to understand how the work itself was changing by leveraging our in-house UX research expertise to study our own team.

Our research drew on several methods to get a 360-degree view of the content workflow. We conducted diary-style surveys, where writers tracked their daily tasks to identify where AI could actually save time. We used group workflow mapping to visualize the journey from a product requirement document to a published help article. To test our content externally, we also conducted a performance analysis of actual Agentforce conversations to see where the AI successfully retrieved our documentation and where it hit a wall. 

What the research actually showed 

Our early findings had nothing to do with AI at all. One of the most significant findings involves an “information gap.” Content professionals have a unique superpower: the ability to synthesize complex organizational context into clear knowledge. However, our research showed that writers often couldn’t lean into this superpower because the raw information they needed, such as technical specs and product insights, was trapped in other departments.

To address this, we’re shifting our focus to making these product insights easier to surface. Our goal is to use AI and automation to turn that raw product information into auto-generated first drafts of help articles and release notes, streamlining the entire workflow. This shift highlights a paradigm change in the role. The writer’s primary value is no longer gathering and assembling information. It’s discernment.

Here’s how we envision this working: When a new feature launches, product managers typically document technical specifications in internal wikis and Slack threads. Previously, a writer would spend hours hunting down this scattered information, translating technical jargon, and drafting an article from scratch. With AI automation, we can pull that information and generate a structured first draft. The writer’s role now shifts to what they do best — determining which technical details matter most to customers in high-stakes moments, refining the voice, and ensuring accuracy. An AI can generate a thousand variations of a sentence, but it cannot make those judgment calls about what matters and why.

One of the biggest risks to our content quality isn’t the AI tools themselves. Instead, it’s the tension between the speed of AI adoption and the awareness of what makes writers in tech companies crucial. They translate, they advocate for clarity, and they uplift what’s most important. 

The messy middle: Balancing innovation with alignment 

Leaning into our superpower requires more than just better information; it requires a coordinated environment. As we pushed for evolution, we encountered a “messy middle.” Our internal research across global teams found that while there was a massive appetite for innovation, the absence of standardized processes created friction. 

• Duplication of effort: Without a central framework, multiple sub-teams often solved the same problems simultaneously. 
• Misplaced innovation energy: Some writers felt a self-imposed pressure to build new AI tools, rather than using existing ones to deliver the core content our customers need. Teams were spending time reinventing solutions for problems that already had functional tools.

Part of our upskilling strategy wasn’t just about showing people how to use new AI tools; it was also about operationalizing innovation. We created a formal “CX AI innovation submission form” to track and vet new ideas, helping us distinguish between problems that needed custom solutions and those that existing tools could already solve. This centralized process ensured that creative energy was funneled toward genuinely unsolved problems, keeping the team aligned rather than fragmented.

How quality and craft are evolving 

As the writer’s role shifts toward discernment and alignment, the way we measure “quality” must also transform. And as the primary consumer of content shifts toward AI agents, our signals for quality must shift as well. Dean Atchison, our VP of Content Experience, notes that while direct ratings and page views are still relevant, they are no longer the only metrics that matter. Instead, we navigate a new set of signals:

• Disambiguation failures: These occur when an agent has to ask a customer, “Did you mean X or Y?” because the underlying content is too overlapping or poorly structured.
• Escalation frequency: These occur when an AI agent reaches its limit and must hand off to a human expert, signaling gaps in our content or moments where customer trust is at risk.            
• Product name changes: In a retrieval-based system, a simple name change can break the logic of an AI agent if the content is not updated across every touchpoint.         

Dean maintains that while these signals are new, our core standards are non-negotiable. Product content still receives a human review by a technical subject matter expert. We also analyze Agentforce conversation data to understand where our documentation successfully supports the agent and where it falls short, allowing us to refine content before issues reach customers. This ensures that efficiency never comes at the expense of accuracy.

Leading the transformation 

Evolving a team of this scale requires a framework for responsible integration. In Sabrina Kang’s research with other tech practitioners, she’s explored a model called the 4C Cycle (Context, Competence, Co-Creation, and Continuous Oversight). This framework, informed by the Stanford Ethics, Technology, and Public Policy course, helps move practice from efficiency-first metrics to effectiveness-first approaches.

At Salesforce, we are seeing these principles in action as we invest in our team’s future. We aren’t just upskilling writers; we are creating new roles, including knowledge architects and LLM specialists. Our knowledge base, which now spans more than 200,000 files, is being intentionally structured for AI retrieval. 

Jo Ward views this shift as an evolution toward governance as a core strategic asset. AI can generate content, but a team has to govern that knowledge layer so that it stays accurate and complete and protects the brand. Ungoverned content degrades quietly. If an agent acts on outdated documentation, the customer loses trust in the product. As Jo puts it, “We must become subject matter experts and content architects.” By positioning our team as the stewards of this infrastructure, we ensure that AI remains a reliable partner, rather than a liability. 

What other content teams can take from this 

The agentic era doesn’t eliminate the need for content professionals. It increases the importance of their judgment. For organizations building AI-powered experiences, the question is no longer whether content matters. It’s whether your team is empowered to govern it. 

Based on our journey, we suggest four pillars for navigating this transition: 

1. Treat content as AI infrastructure. If AI agents rely on your knowledge base, content quality is operationally critical. It’s the foundation of the agent’s performance. 
2. Start with workflows, not tools. Solve the “information gap” first, so your team can use AI to automate the routine and focus on high-impact synthesis. 
3. Governance is the new value-add. Content teams are moving into a new role, from producing at volume to serving as stewards of trust and quality. This ownership cannot be offloaded to product managers, developers, or support engineers without losing the user’s journey. You may end up paying more for reactive content versus a proactive strategy grounded in customer needs. 
4. Operationalize for consistency. Build frameworks that allow for innovation without sacrificing team-wide alignment. Intentionality prevents the “messy middle” from becoming permanent chaos. 

By the numbers, the transformation is clear. More than 3.6 million Agentforce conversations (as of April 2026) and a 60% resolution rate demonstrate the content team’s value in building trusted, AI-powered support grounded in human expertise.