How to deploy the same skill to every rep's AI agent
Updated
The short answer. To give every rep the same skill, decide skill by skill what is locked and what learns: locked skills (the handoff to customer success, lead scoring, what gets written back after a call, discount rules) run from one source that every rep's copy follows, and personal skills, like how each rep writes email, start from a shared template and learn from the rep who uses them. Day AI gives every rep an agent, in place of their CRM or on top of Salesforce or HubSpot, and lets an admin publish each skill from one Skills Library as Managed, where every rep's copy follows the source, or as a Template each rep makes their own.
Say you run customer success at a developer-platform company, and a call with one of your accounts has just ended at 2:10 on a Tuesday. By about 2:40 the follow-up email is in your drafts, written the way you write it, in your own voice. In Slack, your agent has proposed Salesforce changes from the same call: updated MEDDPICC fields on the opportunity, and the people who were on the call attached to it. You approve the Salesforce changes from Slack before your next meeting starts, and the opportunity in Salesforce now shows who was in the room and where the deal stands. The other CSMs on your team get the same kind of proposal after their calls, approved the same way. None of them gets your email. The field update is the part every CSM should do the same way. The email is the part that should sound like you.
Where the usual answer goes wrong. Most advice on sharing a skill stops at distribution: commit the file to a shared repo, upload it for the whole organization, tell everyone to install it. That gets one copy onto every machine and leaves you with a bad choice: lock the skill so nobody can change it, or let each rep edit a copy that stops receiving your fixes. Day AI's position is that agents do the work and people approve it, and that a team's agents need a fixed core and a personal layer at the same time. Day AI provides the context graph and agent control plane to quickly and securely upgrade your company brain, and deciding what sits on each side of that line is most of the work. If you already have a folder of skills that works for you, that was the right start, and it holds until the second rep needs the same skills.
The lock-or-learn split, with Day AI
Christopher O'Donnell, Day AI's CEO, argues that locking everything turns an agent into a script, and locking nothing lets every rep's agent drift. The lock-or-learn split is a decision you make once per skill, for each kind of agent, before you deploy anything. Day AI has a setting for each answer.
- List the skills. Write down every skill your reps' agents run or will run. In Day AI they live in the Skills Library on the Agents page, and tags group them by team, workflow or stage.
- Ask the lock question. If two reps did this differently, would a customer, a forecast or another team notice? Then lock it: the handoff to customer success, lead scoring, what gets written back to the CRM after a call, discount rules, when a deal may change stage. In Day AI, a locked skill is a Managed skill.
- Ask the learn question. Does this get better the more it knows one person? Then let it learn: how they write email, when their morning briefing arrives, how they like a meeting prepped, which accounts come first. In Day AI, a learning skill is a Template skill, a skill the rep writes on their own agent, or a standing preference in the agent's memory.
- Split the mixed ones. A follow-up skill can hold both: the CRM fields it fills are locked, the email's tone is personal. Put the shared rules where one edit reaches everyone. In Day AI, a skill can attach a page it reads every time it runs, so the discount rules can live on one page shared with the workspace.
- Move the line on purpose. Review the split every month or so. A personal skill that several reps have improved may belong to everyone; a rule reps keep correcting may need to be locked. In Day AI, a rep's skill can be added to the Skills Library and an admin can deploy it to the team.
| Skill | Lock or learn | In Day AI |
|---|---|---|
| Handoff to customer success when a deal closes | Lock | Managed skill, deployed to every AE's agent |
| Lead scoring | Lock | Managed skill |
| What gets written back to the CRM after a call | Lock | Managed skill, started in learning mode; CRM write tools set to Needs approval |
| Discount rules | Lock | One workspace page, attached to the skills that quote prices |
| Everyday email in the rep's voice | Learn | The Email Writing skill each agent builds from its rep's sent mail |
| Morning briefing format and timing | Learn | Template skill each rep adjusts |
| Meeting prep | Learn, with locked parts | Template skill, with the shared qualification rules on an attached page |
How to deploy a skill to every rep with Day AI
The steps below are done in Day AI, in the app or from Claude through the Day AI MCP.
Step 1: Give every rep an agent in Day AI
Each rep gets an agent with a name, a job and skills, and the agent reports to them. Each person connects Gmail and Google Calendar, and their calls come in through Day AI's notetaker, Gong or Granola. If you keep Salesforce or HubSpot, connect it through its MCP server with the write tools turned on, so agents can read from it and write updates back. An agent inherits the permissions of the person it works for, and each person sets their own email sharing rules, so the same skill on two reps' agents reads what each of those reps may see.
Step 2: Publish the locked skills as Managed skills in Day AI
Open Agents, switch to the Skills tab (admins also find it under Settings > Agents > Skills Library), create the skill and choose Managed. Only workspace admins can create, edit or deploy a Managed skill. Then click Manage deployment and check the agents that should run it. The review table lists each agent's owner, tier and status before you save, and for a skill that runs on a schedule or an event it also shows how many automated skill slots each agent has left. Every agent you pick gets a synced copy that follows the source. A Managed skill runs with what is shared with the workspace and can't reach any one rep's private email or recordings, so a skill the whole team runs never reads one person's private mail.
Step 3: Publish the personal skills as Template skills in Day AI
For the learn side, create the skill as a Template. Any paid member can publish one. Reps add it from their agent's Browse skills view, or an admin deploys it, and each agent gets its own copy to change. Later edits to the template don't touch existing copies; they reach the next people who add it. Some personal skills need no setup at all: every Day AI agent gets an Email Writing skill once its rep's email has synced, which reads that rep's sent mail to learn their length, tone, greetings and sign-offs.
Step 4: Let the personal layer learn in Day AI
Reps teach their agents by replying. When a rep reacts to a skill's output with a correction or a "always do it this way", the agent writes that into the skill right away, tells the rep what it changed, and keeps those rules in a Learned Preferences section at the end of the skill that the rep can review or remove in the skill editor. Each agent also keeps a memory of its person's standing rules and preferences and follows it in every task. When the feedback lands on a Managed skill, the agent doesn't change it. It tells the rep a workspace admin has to apply it and spells out the exact change to pass along.
Step 5: Push one change to every rep from Day AI
To change a locked skill for the whole team, an admin edits the Managed skill once, and every synced copy follows. Every skill prompt in Day AI is versioned: you can see each saved version, with its author and date, and restore an earlier one. Every skill also keeps a history of its runs, including whether each run delivered its result by email or Slack, so you can check what the new version produced. For a new locked skill that acts on customers or the CRM, start it in learning mode: it reports what it would have done and asks for feedback before it is allowed to act.
Step 6: Run it from Claude with the Day AI MCP
If your skills already live in a Claude Code folder, keep writing them there. Day AI exposes its tools over MCP at https://day.ai/api/mcp (setup for Claude, Claude Code and Cursor is at day.ai/mcp). An admin connected to it can ask Claude to create a skill in the Skills Library as Managed or Template, deploy it to named teammates' agents, undeploy it, or pull up its earlier versions. Day AI's open-source GTM Brain is a Claude Code starting point for planning and configuring a Day AI workspace; the video on this page starts from it.
A 30-day rollout with Day AI
Week 1: Agents and the split in Day AI
Give every rep a Day AI agent and connect their email, calendar and calls. With the team leads, sort every skill you plan to ship into lock or learn, using the two questions above.
Week 2: The locked core in Day AI
Publish the locked skills as Managed skills and deploy them to one team first. Start the ones that act on customers or write to the CRM in learning mode, and read what they say they would have done.
Week 3: The personal layer in Day AI
Publish the learn-side skills as Templates and have each rep on the pilot team add them. Ask reps to correct their agents by replying, and check that the corrections show up in each skill's Learned Preferences.
Week 4: Everyone, and the first change, in Day AI
Deploy the Managed skills to every agent. For a skill that runs on a schedule or an event, the review table shows which agents have a free automated skill slot and a compatible schedule, so free a slot on any agent that needs one before you save. Then ship one real change to a locked skill and confirm in the run history that every rep's agent ran the new version. Add the best personal skills from the pilot to the library.
Where the usual way breaks
- One copy, or many that drift. Shared through a Claude organization, a skill is either view-only for the people it's shared with or a copy each person edits on their own (Claude Help Center). A sales team needs locked stage rules and a rep's own voice inside the same workflow. In Day AI, each skill is Managed or Template, and the rep's personal layer learns on top.
- Nobody knows who is running what. Skills provisioned to a whole organization are on by default, and each person can switch one off for themselves (same source). Add new hires who joined after the rollout, and you can't tell which reps run which version. In Day AI, each skill's deployment list shows which agents have it, with each agent's owner.
- It runs only when the rep remembers. A skill in a chat session runs when someone asks for it, so the follow-up rules apply only on the days a rep thinks to use them. In Day AI, skills run on a schedule, when a recording is ready or an email lands, or on request, for every rep's agent.
- Each copy sees only its own runner's data. A skill reads what the person running it has connected, and pooling everyone's mail into a shared folder exposes it to everyone with access. In Day AI, every agent reads the same customer memory, limited to what its person may see.
- The good edits never come back. When a rep improves their copy, the improvement stays on their laptop, and the rest of the team never gets it. In Day AI, a rep's skill can be added to the Skills Library, and an admin can deploy it to the whole team.