Salesforce: AI Skills Turn Coworker Instructions Into Reusable Workflows

Salesforce says AI Skills in Agentforce Coworker are now generally available, according to a company blog post. The feature lets users define a reusable set of instructions that tell an agent how to complete a repeatable task: a job described once in plain language that Coworker can then run consistently. Salesforce describes Coworker as an AI teammate that coordinates Agentforce agents to complete complex tasks in Salesforce, Microsoft Teams, and on mobile, and says it is the first Agentforce agent to include AI Skills.

Salesforce frames the launch around expertise held by individuals rather than spread across a team, citing the way a top-performing sales representative preps for a first call or the checklist a manager runs before a forecast review. Those workflows are hard to scale, the company says, because nobody has written them down in a form AI can run.

Key Takeaways

• Salesforce states that AI Skills in Agentforce Coworker are generally available, and that Coworker is the first Agentforce agent to include them.
• A skill describes an entire repeatable task so an agent can work through each step and take action, which Salesforce contrasts with a Prompt Builder prompt template that produces text outputs such as an introductory sales email or a record summary.
• Skills are private by default until the creator shares them, and a skill runs as the person using it, with that person's row-level and object-level permissions.
• Salesforce says Coworker pauses for approval before it creates or updates a record.
• Users can invoke a skill by typing a forward slash, clicking the skills icon in the chat box, or asking Coworker, which can also select a skill automatically based on intent.

Context & Analysis

According to Salesforce, a skill begins as a conversation in Coworker, which users open from Salesforce by clicking Ask at the end of the search bar; saved skills sit under Skills in the left sidebar. Building one takes three steps: describe the workflow, review and refine the steps in Coworker or open them in Skill Builder to edit directly, then save and share. When creating a skill, Coworker asks the user to name it, explain when to use it, and list the steps it should follow, such as which records to check and how to format a meeting brief.

The contrast with prompt templates is central to the pitch. Salesforce describes a prompt template as a repeatable way to prompt a large language model for text outputs, whereas a skill describes the whole task so an agent can execute each step and act. That distinction matters for governance. Because action-oriented behavior runs inside the existing permission model and pauses for approval before record changes, Salesforce presents skills as staying within guardrails an organization already maintains rather than reaching data the running user could not already see.

The post quotes Avanthika Ramesh, Salesforce's Senior Director of Employee AI Workflow Products, calling AI Skills a paradigm shift in how work gets done and describing the goal as turning isolated expertise into collective, scalable intelligence with built-in trust and governance. Salesforce lists five illustrative task areas: account research; call prep and record sync; competitive battle cards; case summarization and response drafting; and turning campaign performance into a report. These are presented as ideas teams could adopt, not as measured outcomes.

Competitive Landscape

The verified evidence supports comparison only among components Salesforce itself describes inside its platform. The table compares internal approaches and roles, not the wider market.

What It Means

For Enterprise Buyers

Teams evaluating AI Skills should treat it as a way to standardize recurring processes rather than as a general-purpose assistant. Because a skill runs under the invoking user's row-level and object-level permissions and Coworker pauses for approval before creating or updating records, administrators can assess exposure using controls the organization already maintains. The practical diligence question is which repeatable workflows, such as account research or case summarization, are documented well enough to encode, and who should own sharing them.

For Investors

The signal is incremental rather than financial. Salesforce is positioning skills as a governance-friendly layer on top of Agentforce and its prompting tools, with sharing and permissions framed as adoption enablers. Investors should seek customer counts, revenue figures and usage data before drawing conclusions about monetization, attach rates or productivity gains.

Forward Outlook

Documented milestones are narrow: general availability is stated, and Salesforce points readers to a demo video and an AI Skills help article for further detail, alongside an AI Fluency Playbook for leaders. Unanswered questions include how organizations will govern a growing library of shared skills, how approval pauses affect longer workflows in practice, and whether the five example tasks perform as intended across varied data. Buyers should verify licensing, pricing and access terms directly before deployment. Salesforce's advice to start with one weekly workflow and expand after validation is guidance rather than evidence of outcomes.

Direct source: Salesforce blog post.