From copilot to colleague: Agentforce transforms work with autonomous AI

Last September, Salesforce launched an exciting new chapter in the book of AI: Agentforce, AI agents that can reason and perform tasks independently. While traditional chatbots and copilots still rely significantly on human intervention, AI agents function more like digital colleagues who proactively and autonomously complete tasks.        

Since 2014, we’ve pioneered enterprise AI, starting with Einstein Predictive AI. And we didn’t slow down. We brought generative AI to our customers as a next wave. Building on that momentum we focused on our platform, started building features that enhance the security and efficiency of AI and reducing barriers to entry at the same time. For example, the Einstein Trust Layer, which is part of the Salesforce platform, securely connects customer data to Large Language Models such as OpenAI, preventing that data from being used in training models or revealing identifiable information.

The following step introduced a generative AI agent, Einstein Copilot, capable of executing actions based on user instructions. Together with 100 customers, we launched pilot projects that allowed us to gather learnings and refine the technology. This resulted in our new reasoning engine, “Atlas,” that enables AI to plan, breakdown and execute specific tasks more autonomously.

Agentforce is the latest evolution of Einstein Copilot, giving users the opportunity to support their work with both a digital assistant and an autonomous agent. AI can now really do the work for you – or as our CEO Marc Benioff expressed it at Dreamforce: this is what AI is meant to be.

How AI agents work

Traditional bots required extensive programming with fixed scripts and structures. For example, creating bots in multiple languages meant building separate bots for each language. With Agentforce, AI agents or agentic AI use natural language, reducing the need for hard coding. Users assign a role to an agent – like a first-line support representative – and define accessible information from both internal and external sources. This includes structured data (e.g., order details) and unstructured data (e.g., prior customer interactions, knowledge articles, company policies, etc.).

Agents can then be assigned actions linked to specific topics, like opening a case or initiating an order return. You can also set guardrails on actions they are restricted from – such as allowing returns on orders of less than 100 euros. The agents will break down every question and engage a human employee if they are unable to respond or if the use case is outside the predefined guardrails.

Making your workforce more flexible

Agents increase workforce elasticity and flexibility. Especially during high-demand periods like Black Friday or the holiday season, they reduce pressure on human employees. Rather than eliminating jobs, they eliminate tasks by automating routine work.

Recently, as part of Agentforce, we launched autonomous Service Agents. As contact centers face staffing shortages, AI can handle routine queries, such as password resets and order status checks. Service Agent assists contact center teams with tasks like responding to WhatsApp or Facebook Messenger messages, chat on websites or mobile applications… Soon, Service Agent will even extend to emails and phone calls.

The upcoming Sales Agent will support sales teams by assessing website leads, conducting initial outreach, and further qualifying the leads. If an opportunity emerges, the agent can automatically schedule a meeting with a sales contact. This feature enables sales teams to focus on high-potential leads from the outset.

Outpacing copilots

AI agents are wave 3 in AI. They surpass copilots in autonomy, security, and precision. For example, Agoria and the Dutch Data Protection Authority (DPA) have recently raised concerns about data leaks with copilot models. Salesforce’s AI agents, in contrast, use other techniques such as Retrieval Augmented Generation to enrich prompts with business data, while keeping existing security permissions and policies intact. This approach ensures that users see only authorized information, with all outputs verified for security via the Einstein Trust Layer.

For customers, AI agents maximize the value of prior investments as they seamlessly connect to data and workflows already within Salesforce. Soon, every employee or team will have a digital assistant or colleague. The manager of the future will lead a hybrid team of both human and digital employees. However, this is by no means a distant future. Within the next five years, the ability for people to collaborate with AI agents, as well as AI agents with other AI agents, will redefine our ways of working and create endless possibilities. Far beyond what copilots were designed to deliver.

Upcoming blog posts: learn about the future of AI agents

Explaining the magnitude of this revolution is impossible in one blog. Together with my colleagues, we will share a series of blog posts in the upcoming months. Our experts will draw a picture of the potential of our AI agents in sales, service and marketing teams. They will focus on specific industries, such as finance, manufacturing, retail, and telco. To explain how AI agents work from a technical perspective, we will also look under the hood of this exciting new technology.

Stay tuned for more information! If you already have a question or a comment you would like to share, make sure to send me a message.