According to a publication by Salesforce, the AI capabilities known as AI Skills within the AI agent Agentforce Coworker are now generally available. According to the company, this represents a new way for users to define a reusable set of instructions that guides an agent on how to complete a repeatable task. The user describes the work once in plain language, and Coworker can run it consistently every time. Salesforce describes Coworker as an AI teammate that coordinates Agentforce agents to complete complex tasks across Salesforce, Microsoft Teams, and mobile devices, and it is the first Agentforce agent to include AI Skills.
What is the Difference Between AI Skills and Prompt Templates
The publication explains that every team has best practices that reside in the minds of specific employees. Examples provided include the way top-performing sales representatives prepare for a first call, the checklist a manager runs prior to a forecast review, and the exact steps for entering meeting notes into the correct record. For the people who know them, these workflows represent a significant change, but they are difficult to scale across the organization because no one has written them down in a format that artificial intelligence can execute.
The publication compares the new capability to Prompt Builder, which features prompt templates grounded in the organization's Salesforce data. Unlike a prompt template, which provides a repeatable way to prompt a large language model (LLM) for text outputs—such as an introductory sales email or a record summary—a skill describes the entire repeatable task. This description enables the agent to work through each step of the task and take actual actions.
In addition, it is noted that every team has seasoned employees who know the shortcuts, the checks, and the steps that are easy to miss. Using AI Skills, these employees can convert that process into a skill by holding a conversation with Coworker. Afterward, anyone with whom the skill is shared can run the exact same process consistently through Coworker at any time, including new hires or teammates.
Avanthika Ramesh, Senior Director of Employee AI Workflow Products at Salesforce, noted that AI Skills mark a true paradigm shift in how work gets done. According to her, empowering every employee to codify their routine workflows into agentic skills and share them turns isolated expertise into collective, scalable intelligence across the entire company, alongside the built-in trust and governance teams need to move fast.
How to Build a Skill and Run It in the Workspace
The process of creating a skill begins as a conversation inside Coworker. To open Coworker, a user logs in to Salesforce and clicks "Ask" located at the end of the search bar. Coworker opens with a box where a request can be typed, alongside a suggested question to assist in getting started. The user's skills are located under the "Skills" tab in the left sidebar.
Building a skill comprises three steps:
- Describe the workflow: The user opens Skills in the Coworker sidebar and clicks the New button. In response, Coworker opens a chat that walks the user through the process, asking them to name the skill, explain when it should be used, and list the steps it needs to follow, such as which records to inspect and how to format the meeting brief.
- Review and refine the steps within Coworker, or open them in the Skill Builder editor to edit them directly.
- Save and share the skill: Skills are set to private by default, meaning the skill remains private until the user chooses to share it with the individuals who need it.
Regarding running a skill, the publication notes that the process is simple and offers several options: typing a forward slash (/), clicking the skills icon in the chat box, or simply asking Coworker for the required action. Coworker is also capable of selecting the appropriate skill automatically based on user intent, so a sales representative can ask for meeting preparation without needing to remember what the skill is named.
Furthermore, skills operate within the guardrails that already exist in the organization. A skill executes under the identity of the user running it, in accordance with their row-level and object-level permissions, so it cannot access data that the employee would not have been authorized to view anyway. Additionally, Coworker pauses and waits for approval before creating a new record or updating an existing record.
Five Tasks That Can Be Handed Off to an Agent-Based Skill
Salesforce notes that skills are suitable for most tasks that recur within a team, and presents five examples of their application:
- Account research: Researching an account typically requires opening several different tabs, such as the account record, recent news, open opportunities, and past activity. With a skill, the team writes down its research routine once—including which records to examine first and what data should be highlighted in every summary. Subsequently, any representative can ask Coworker to research an account and receive a summary constructed in the exact same format, ensuring all representatives operate from the same baseline.
- Call prep and record sync: Prior to a customer conversation, Coworker can follow a skill to assemble the context needed by the representative and generate a brief in a standardized format. Following the call, it can synchronize follow-up actions back to the record once the representative confirms. In this manner, the prep routine of the top representative becomes the routine of the entire team.
- Competitive battle cards: Instead of representatives searching on their own for how to position against a competitor, a battle card skill can gather talking points, differentiators, and objection handling into a single consistent response right when the representative needs it.
- Summarize a case and draft a response: When a service case involves a tangled history dispersed across emails, chat conversations, and notes, a skill can summarize all the information and compose a customer-ready draft response in a single step. The representative reviews the draft, makes adjustments if necessary, and sends it.
- Turn campaign performance into a report: A campaign performance skill collects the latest data, summarizes what is functioning well and what is not, and presents an organized, ready-to-share report, replacing hours of data exporting and formatting in a single run.
How to Get Started with Skills
The publication recommends that teams interested in experimenting with AI Skills initially select one workflow performed on a weekly basis, and write down its steps in the same manner they would explain them to a newly hired employee joining the organization. Next, they should open Coworker, guide it through those steps, and share the skill with teammates once it operates according to their preferences.
For further learning, the company directs users to view a demo video and bookmark the help article covering AI Skills. Finally, the AI Fluency Playbook is noted, which shows leaders how to build AI engagement, activation, and expertise.