How Does the Agentic Enterprise Really Work? Lessons from Year One

Stephanie Losee had been a people manager at Salesforce for three months when she realized she was about to miss a critical deadline. Employee performance evaluations were due in 48 hours, not the six weeks she thought she had. She panicked. At her previous job, she had spent as much as 20 hours per employee gathering feedback, writing reports, and comparing performance levels. With five direct reports who qualified for calibration, that would have meant 100 hours of work compressed into two days.

At Salesforce, she had a different option with Slackbot and Agentforce. Opening up Slackbot, she asked it to pull a year’s worth of feedback for each employee from documents and Slack channels, including conversations that happened before Losee joined the company. 

Stephanie then went into the company’s performance evaluation tool, where the Manager Agent drafted performance summaries, which she edited and supplemented with the feedback that Slackbot had surfaced. In eight hours total, she completed what would have taken her weeks.

“The amount of panic I had was way out of proportion with what the actual task ended up being,” Losee said, “because of all the support that the AI agents give managers at Salesforce.”

Three weeks later, looking back at what the agent had generated after gathering more information, Losee realized something else: It had been right on the money. She’d barely needed to tweak it. This is what Salesforce calls living in the Agentic Enterprise.

An Agentic Enterprise is a company where humans and agents collaborate seamlessly. Every employee is augmented, every decision is guided by trusted data, and every customer experience feels like one continuous conversation. At Salesforce, that vision is powered by Agentforce, the company’s AI agent platform, integrated with Slack as the central workspace where all communication and collaboration happens.

So how does it actually feel to work in one?

At Salesforce, the answer involves a lot of trial and error. The company has spent the past year transforming itself into “Customer Zero” for the Agentic Enterprise by building AI agents internally, prototyping the path for customers who seek to do the same. That experimentation has crystallized into a mission statement, as Salesforce President Enterprise & AI Technology Joe Inzerillo, who oversees Salesforce’s Agentic Enterprise transformation, put it: “Humans for impact, agents for scale.”

From Email Jail to Strategic Work

For Losee, working in an Agentic Enterprise meant immediate escape from what employees around the world call “email jail.”

At her previous job, she would process emails over holiday breaks just to avoid returning to 5,000 messages on January 2. “Email jail was a huge issue in my previous work life,” Losee said. “I felt a fair amount of daily shame.”

But at Salesforce, it’s different. “I have over 1,500 emails,” she said, “and I don’t feel like I have to get to the bottom of them.” That’s because at Salesforce, work happens in Slack. She scans her inbox daily in case something comes in from an external vendor, but even most agencies work in Slack channels now.

Every morning, Losee asks Slackbot: “What should I know about in my Slack channels? What in my Slack channels should I prioritize today?”

It surfaces the deadlines, decisions, and conversations that need her attention. While Losee says it’s not perfect, it’s so close to the mark that it organizes her life as a manager.

“I’m now spending almost all of my time on strategic tasks,” Losee said. “I do so little running around trying to process emails and complete these low-level tasks.”

It’s a pattern playing out across Salesforce. The HR Support Agent has already saved 3.6K annual hours by reviewing cases and proposing solutions. The Techforce Agent resolves 35% of IT issues independently. In the last promotion cycle, the Manager Agent handled the process and paperwork for 95% of promotions, though the actual decisions about who gets promoted still belong to humans.

Eighty-five percent of Salesforce employees now feel confident using AI tools in their roles. Translating that confidence into actual business impact is Giulia Sergi, who leads AI fluency strategy for Salesforce’s Workforce Innovation team.

“We’re seeing the usage. We’re seeing the adoption,” Sergi said. “What we need to do now is translate that activity into impact.”

That’s AI fluency — the ability to confidently collaborate with AI to give employees agency and drive business impact at speed and scale. But getting there requires navigating what Sergi calls the “messy middle.”

The biggest barrier isn’t technical. It’s trust, which manifests in several ways. Employees try an agent, and if it doesn’t give them what they expected, they give up. There’s “incompetency bias,” the fear that being transparent about using AI means admitting you can’t do the work yourself. And there’s the “verification tax,” spending extra time validating AI outputs because the trust isn’t fully there.

Sergi tells employees to think about agents like interns joining their team. You wouldn’t expect an intern to know everything on day one. “It’s the same thing with an AI tool or an agent,” she said. “How can we support them and make them more effective? We have to rely on our business team of experts to help test and fine-tune the results.”

Salesforce is tackling the trust issue by making AI use routine and visible. Managers on some teams have begun to share weekly what they tried, what worked, and what didn’t. Teams set aside 10 minutes in meetings for social learning and showing each other how they’re actually using the tools. The approach is working: When managers lead by example, teams engage with AI tools 25% more, creating what Salesforce calls “The Leadership Multiplier.”

“We’ve found that that is the number one driver of AI adoption,” Sergi said, “seeing other people that do similar jobs to yours use the tools and learning how they’re using it.”

You can get something up and running, but the real value and learning starts when you go live and you see how it’s being used.

Andy White, SVP of Digital Enterprise Technology at Salesforce

Putting Agents to Work

Social learning drives adoption, but business pain accelerates it. In Salesforce’s Sales organization, sellers were using over 14 different tools with inconsistent approaches to create a point of view for an account, according to Samantha Maurovich, who leads change management for Sales. Some relied on manual offline tools, others used internal AI workflows or external AI research assistants, and still others built custom solutions. 

The best sellers had a process that involved deep research on the prospect, understanding their challenges, and crafting a specific pitch. That work took hours per account, which meant only the most diligent sellers could deliver high-quality POVs consistently. Everyone else cobbled together what they could.

The Account POV powered by Sales Agent was purpose-built to break that constraint. It does what top performers do (research, synthesis, tailored outreach) but in minutes rather than hours. Account POV ensures every seller can deliver a polished, credible, and customer-ready point of view with one click. The payoff for this functionality, and others in the Sales Agent, is over 200,000 hours saved annually and more time spent with customers.

Once a seller has a qualified meeting scheduled, Agentforce’s Sales Coach adapts to the deal stage. For opportunities in qualification or needs analysis, it analyzes the seller’s pitch and provides targeted feedback. For deals in negotiation or pricing, the agent role-plays as the customer, letting sellers practice handling objections and pricing conversations before the real thing. The agent cross-references internal opportunity data to make the coaching specific to each deal, providing personalized feedback on everything from value proposition strength to delivery style. The result is a 10% increase in win rate, a significant gain for a mature sales organization.

“Before this, sellers were practicing in the mirror or relying on their manager’s capacity,” Maurovich said. “Now they’ve got an always-on coach available anytime, which is huge for new hires who need the repetitions and guidance without worrying about looking unprepared or pulling their manager into every question.”

Getting sellers to actually use the agents required a three-pronged approach. Maurovich built a global champion network of early adopters who wanted to test the tools and report back what worked and what didn’t. Like Sergi, she also focused on leader-led learning, getting sales managers to try the agents themselves and share their experiences. 

She scaled the approach with “AI in Action” roadshows in nine global hubs. She took a genius bar approach, where sellers could walk up, see the agents in action, and try them out. The feedback didn’t stop after rollout. Salesforce built open loops where sellers can flag bugs, request new capabilities, or share success stories. “We know we’re not done there. In fact, we are just getting started,” Maurovich said.

Learning What Actually Works

Building ​​agents that employees will use and trust required Salesforce to learn some hard lessons first. When Agentforce launched, teams across the company built more than 200 agents. It was what Andy White, SVP of Digital Enterprise Technology at Salesforce, calls the “kids in a candy store” phase.

“We started out not really understanding the level of investment that it is to make an agent great,” White said. “You can get something up and running, but the real value and learning starts when you go live and you see how it’s being used.”

The first surprise came when White’s team launched an agent on Salesforce Help. Customers asked different questions than what was predicted, like developer questions. They hadn’t expected developers to want to use an agent to answer these questions, so developer documentation wasn’t one of the initial data types they initially hydrated.

Two months in, White’s team realized the answers the agent was giving felt cold. They’d onboarded all the technical documentation and knowledge articles but missed something critical. “We didn’t train it at all on the art of service,” White said.

Now Salesforce thinks about agents in terms of “head and heart.” The head is technical knowledge — the facts, the procedures, the documentation. The heart is the experience the customers have — the tone, the approach. The Lead Engagement Agent, which works sales leads 24/7, needed the heart of a seller: hungry, persistent, always looking for the deal. The Help Agent needed the heart of a service professional: trustworthy, patient, empathetic, and focused on a resolution.

Getting the soft skills right transformed the agents from answer machines into something that felt like what people have come to expect from Salesforce’s support experiences. The Lead Engagement Agent now follows up on every lead, not just top-tier prospects. Human sales development reps wake up to full schedules with prequalified conversations. White calls it “sawdust turning into gold” — revenue from those 100 million leads that used to sit untouched.

Agents still struggled with something humans do instinctively: filtering and curating. White’s team thought the Help Agent was hallucinating, but upon a deeper look, it came back to the source material. Password reset documentation covered multiple platforms — Tableau, Sales Cloud, etc. The agent was pulling information from all of them at once. The solution was to instruct the agent to seek clarification before providing an answer. To catch problems before they reach employees or customers, Salesforce has implemented an “Agents Testing Agents” system that uses AI to evaluate Agentforce’s responses based on relevance, correctness, and completeness. 

To overcome the unpredictability of real-world conversations, the team utilizes “Synthetic Utterances” — controlled, realistic customer questions derived from actual support data — to audit performance at scale. This process generates an Answer Quality Score, which allows Evaluation Managers to review failed responses and classify the issue. These specific insights help Salesforce’s engineering and data science teams to investigate and resolve underlying issues.

“You have to monitor what the agents are doing,” White said. “The whole idea is humans for impact and agents for scale. We couldn’t, at scale, have humans look at all of those, but we can have agents do it and tell us where to look.”

What’s Next

Salesforce has now consolidated more than 200 agents to half that number, focusing on the ones doing the heavy lifting. Year two is about scaling what works, according to White. More experimentation, faster iteration, reusable components that teams can share instead of building from scratch. 

The Help Portal has now handled 3 million conversations. The company now provides instant 24/7 fast support in languages it couldn’t staff for before, helping more customers to be successful, even as the customer base grows.

White estimates Salesforce is six to nine months ahead of its customers in figuring out what actually works. That lead time is the entire point of being Customer Zero.

“We make the mistakes. We figure out the fixes. And by the time it gets to our customers, they’re not starting from scratch,” White said. “They’re starting from our lessons learned.”

Those lessons come from listening. Salesforce is constantly collecting employee stories about how agents are changing their daily work, what’s working, and what’s still causing friction, according to Ameeta Ambekar, who leads change enablement and adoption for the Agentic Enterprise. 

Those stories shape what the company calls “the blueprint for the Agentic Enterprise” — proof that the technology works because employees are putting it to use. And with thousands of employees across sales, engineering, finance, support, and dozens of other functions, the technology gets tested from every angle.

“If pressure makes diamonds,” Ambekar said, “Customer Zero is how we pressure-test our products to make sure they’re perfect for customers.” 

Go deeper:

• Learn what it’s like to collaborate with Slackbot, everyone’s new work BFF
• Read how conversation design enables the Agentic Enterprise
• Find Salesforce’s position on where AI stands today: In 2025, AI Grew Up — and Learned to Play by the Rules