The New Collaborative Workforce: Humans and Digital Agents

"Employees, I'd like to introduce you to AI agents. Agents, these are your new partners - our talented human workforce."

Welcome to the new get-acquainted conversation in the workplace, where human workers are collaborating increasingly with AI agents on everything from sales support to marketing programs to engineering projects.

It's a new way of working, one that involves learning and adaptability. Best practices in how to build and scale a digital workforce include having the right agentic platform, infrastructure, policies, and training. Also required: a rethinking of traditional roles.

There's tremendous potential upside in this new world of digital labor - productivity gains, time savings, increased responsiveness - but also challenges. One of the biggest is employee readiness. The human workforce must learn how to interact with their new agentic colleagues - and to trust them.

This may require a degree of cultural adjustment. A Slack survey of 17,000 workers found that nearly half (48%) indicated they would be uncomfortable admitting to their boss that they use AI for common workplace tasks. One way to ease that perception is through exposure to these new technologies, said Lori Castillo Martinez, Executive Vice President of Talent Growth & Development at Salesforce. For example, Salesforce hosts "learning days" where employees can try out new AI tools with colleagues.

It's not just the rank and file. Business leaders also have to adjust. "From this point forward… we will be managing not only human workers but also digital workers," Marc Benioff said earlier this year at the World Economic Forum in Davos, Switzerland.

That vision is fast becoming a reality. Salesforce has introduced Agentforce - a digital labor platform for building and deploying agents that can reason, learn, and act autonomously - for roles in Sales, Service, Marketing, and Commerce, and across industries, including Retail, Health, and Financial Services.

These digital coworkers can be customer-facing or employee-supporting, as illustrated here:

An Agentforce Sales Development Representative can engage customers:

• Send customized outreach emails to prospects using data from the lead or contact record
• Follow up with a nudge email
• Answer questions from prospects
• Hand off interested prospects to sales reps to book a meeting

An Agentforce Sales Coach can support internal development:

• Provide sellers with stage-specific feedback to help advance deals
• Analyze a seller's sales pitch
• Use role-playing to act as a customer for improved interactions
• Cross-reference what the user said with CRM data to land the opportunity

Agents as employee partners

Increasingly, businesses are introducing Agentforce for employee use cases, making it possible for any employee to have a task-specific agent that works alongside them.

Already, agents are appearing in Slack channels where they're part of the team dynamic, the result of integration between Agentforce and Slack. One of the first use cases is using agents to onboard new hires and help them come up to speed quickly.

Case in point: Engine, a modern travel platform, plans to launch agents that provide IT support for new hires and answer role-specific policy questions across the company. For example, if an employee wants to check on the status of their new laptop, they can simply ask Agentforce in Slack, using natural language. Or if a new hire has trouble accessing their email for the first time, they can ask Agentforce for help and instantly receive step-by-step instructions.

These workforce agents are much more than just AI assistants. Think of them as "partners" to employees, said Jayesh Govindarajan, Executive Vice President of Salesforce AI. "The difference," he explained, "is the agent's awareness of the user context, the tailoring of responses and actions based on an employee's role, and the permissions and tasks the agents are given."

With Agentforce for employees, Govindarajan said, you can give an agent instructions to perform a task, and "it goes and gets it done." Agentforce accomplishes this by deriving context from an employee's digital workstream and reasoning to break tasks down. Agentforce then orchestrates actions on behalf of the user, with permission and, if needed, clarifications from that person.

This dynamic of humans and AI agents working hand in digital hand requires supervision. With a hybrid workforce, managers may need to reframe how they gauge teamwork and performance. Are AI agents being utilized as effectively as possible? What can be learned from employee-agent interactions? Adjustments may be necessary to get the synergies right.

A symbiotic relationship

Here's a scenario for what the new collaborative workforce might look like in practice:

Kate, a marketing manager at a consumer goods company, starts her day chatting with an AI agent that has analyzed the performance of an overnight campaign, flagging trends and drafting recommendations before she's even finished her coffee. Throughout the day, the AI agent seamlessly handles routine tasks - scheduling meetings, summarizing documents, mining customer data for insights, and taking actions autonomously. Kate is able to focus more of her time on strategy and relationship building.

In the emerging hybrid workplace, the agent-human relationship is symbiotic, not simply automating rote tasks. This new professional duality combines the power of AI computation with the emotional intelligence, judgment, and innovative thinking that remain uniquely human strengths. AI agents learn on the job, taking in the context from Slack and human interactions to guide their actions.

In the emerging hybrid workplace, the agent-human relationship is symbiotic, not simply automating rote tasks. This new professional duality combines the power of AI computation with the emotional intelligence, judgment, and innovative thinking that remain uniquely human strengths.

Multiply this type of 24×7 productivity by the number of employee-plus-agent partnerships across an enterprise, and you begin to appreciate the ascending business value in digital labor.

Researchers at the Massachusetts Institute of Technology ran an experiment to understand the effect of human-agent collaboration on teamwork, productivity, and performance. Significantly, humans matched with agents enjoyed a 60% productivity boost. Perhaps even more noteworthy: When AI agents with certain "personality traits" were paired with humans, workflow and output were further enhanced, according to the MIT Initiative on the Digital Economy.

Of course, you can't manage what you can't measure. Strategy requires key performance indicators (KPIs) to gauge progress, value, and return on investment. That includes taking the pulse of employee experiences in areas such as job satisfaction and reduced burnout. Other needles on a digital labor dashboard might include customer satisfaction scores, quantitative assessments of tasks completed, and cost savings calculated from agents managing workloads.

Start with value, not technology

Managers must do upfront work to ensure their organization is well prepared for this new style of working and collaboration. Step one is to identify specific business problems - for example, low inventory levels, a need for increased lead generation, or slower-than-desired time to resolution - and value opportunities that can benefit from human-AI teamwork. Then map complementary strengths where AI agents take on routine, data-intensive, and always-on tasks, while human workers do what they do best - exercise judgment, creativity, and relationship management.

Like other AI transformations, this starts with data. The LLMs providing inputs into agents must be grounded in trusted enterprise data, including customer data, telemetry data, emails, audio, and PDFs from a wide variety of sources. Salesforce's Data Cloud, with APIs, connectors, and other integrations to hundreds of sources, provides the needed data connectivity. The Atlas Reasoning Engine, meanwhile, functions as the brain of Agentforce, enabling agents to understand human intent and take actions.

Human employees and agents, with the appropriate access controls, can access and update the same data, ensuring accuracy and consistency in their joint efforts. And that can happen whenever and wherever work gets done because Agentforce works across the surfaces that employees use, including Slack, Tableau, and Customer 360 applications.

Agent design is a top priority. Factors to consider include the roles of both the agent and the human worker, and the data sources required to support their mutual workflows. Govindarajan frames it as user-context awareness. "For these agents to work in the real world, they need to have the right access controls, the right user context, and the right set of actions for the persona," he said.

Policies, principles, guardrails

Design and data privileges are part of the fabric of a well-conceived trust framework that addresses privacy and ethical considerations proactively. This includes bias controls and audits to ensure AI agents behave consistently with organizational policy and ethics.

Agentforce's Trust Layer bakes policy and principles into what AI agents can and can't do. Capabilities include bias and toxicity detectors - like security, privacy, and safety controls - audit trails, and the ability to nudge human intervention, if needed. The Trust Layer also serves as a gateway to LLMs to filter out unwanted behavior or inaccurate outputs, helping to reduce hallucinations.

Managers need visibility into the effectiveness of agent-human interactions. That's where Agentforce Interaction Explorer can help by providing detailed reporting and analytics on how agents are performing, including context, quality, and session tracing. "This tells you whether the employee and the agent are deriving value from each other. It's a powerful construct," said Govindarajan.

Interaction Explorer provides insights into a customer's or employee's request, the reasoning steps associated with an agent's response, and AI-powered recommendations to help refine instructions to improve agent performance. It's like a supervisory dashboard for the agentic system, helping to improve and refine interactions.

Adaptation and continuous learning

Establishing the agentic infrastructure is half the preparation; the other half is getting employees and managers trained and educated for what's ahead. "It's a new skill to be learned," said Govindarajan.

It may involve skills assessment and a transition strategy. Because agents augment human know-how, some roles may need to be redefined. The guiding principle is to do so in a way that leverages uniquely human strengths alongside agentic AI capabilities. For example, a customer service rep might spend more time focused on hard-to-solve customer issues that get escalated. Or a financial advisor might provide the voice of experience during market volatility.

Clear policies and governance can help ensure that human oversight comes in where needed, guiding agentic workflows and setting the right guardrails. Over time, employees will gain experience and confidence with agents, and agents will learn from the interactions, as well. This leads to continuous improvement as digital labor takes root.

In the early going, agent-to-human interactions may be outside the comfort zone of some employees. Employees can overcome their hesitation with a beginner's mindset, where first-time users start small, learn, iterate, and eventually do more. The newly formed Agentblazer Community is an excellent starting point for anyone in need of support. The community can help demystify agentic AI, provide access to technical content and expertise, and foster learning through peer networking.

Managers are integral to facilitating this new style of team building and professional development. "It's what great managers already do," said Govindarajan. "You make sure there's open dialogue with employees and that they're learning from each other."

Now, that same managerial mindset must be brought to bear so that AI agents learn from human workers, and vice versa.

Go deeper: