How the agencies we work with in Canada get the whole team using Cowork, connectors and skills, with the policy, roles, 90-day rollout plan and measures that support it. Every template is copyable.
Tick what’s true for the whole team today, not just for you. Most owners we talk to are the heaviest user in the building, while everyone else is split across personal accounts, other tools and nothing at all.
Team or Enterprise, not personal accounts or whichever tool someone already pays for.
Brief, brand, voice and approved work, loaded once and shared with the account team.
An owner switches them on once; everyone signs in with their own account.
Each skill with an owner and a date someone last checked it still works.
One page, including what client data may go in and who signs off.
With hours in their week for it, not “whoever’s keen.”
A weekly client report or Monday picture that nobody has to start.
Timed by the clock for at least one workflow, not estimated.
The same three bands as the Agency Claude Checklist, which covers the habits for each person. More on each band at kyokolabs.xyz/find-your-starting-point.
Seven clauses and a table. Swap in your names, have it read by whoever handles your contracts, and ask everyone to read it before their seat goes live.
| Data class | For example | In Claude? |
|---|---|---|
| Public | Published copy, press releases, public research | Yes, on any approved surface. |
| Internal | Our processes, templates, rate cards | Yes, on business seats. |
| Client-confidential | Briefs, unreleased work, client analytics and code | Yes, in that client’s Project, once the contract check is done. |
| Personal | Names, emails and phone numbers of your clients’ customers; staff records | Only what the task needs. Never health, HR or payment data. |
Have a lawyer read your version before you adopt it.
Keep the best-practices manual you wrote; it’s a head start, and most of it carries over. What one central tool adds is the layer a manual can’t hold: the same client Projects, connectors and skills for everyone, under one admin and one data policy, so a good prompt becomes a skill the whole team runs instead of a tip in a doc. Move one workflow at a time rather than on a switch-over day, and let people keep what works for them in the meantime.
The most common reason a rollout stalls is that nobody owns the setup. R does the work, A answers for it, C is asked first, I is kept in the loop.
| Decision or task | Owner / sponsor | AI lead | Workflow owners | Everyone |
|---|---|---|---|---|
| Tools, seats and budget | A | R | C | I |
| The AI policy | A | R | C | I, reads it |
| Switching on a connector | A, switches it on | R, writes down access | C | I |
| A new shared skill | I | A, adds it to the registry | R, builds and tests it | C, tries it |
| Client-facing output | I | I | A, named reviewer | R |
| Before and after numbers | I | A | R, times the workflow | C |
Thirty minutes a week: unblock, decide, and turn up to the show-and-tell. People copy what the owner does.
About five hours a week while a workflow is in build: the registry, office hours, the numbers. Pick for curiosity over seniority.
One each for account management, creative, strategy, dev and ops or finance. Each owns the skills in their area and checks them monthly.
This is the part that makes it one team instead of twenty personal habits. Each piece has an owner, and each one is something a new hire inherits on day one rather than rebuilds.
An owner switches each connector on for the org; everyone then signs in with their own account, so Claude only sees what that person can see. The first five: Google Drive, Slack, Gmail, your tracker and Figma.
Write one line per connector on what it can read and what it can change. That note is what lets the owner say yes quickly.
One per client: the brief, brand guidelines as text, tone of voice, approved past work and the instructions that matter. Shared with the people on that account.
The account lead keeps it current. Test a new one with three real asks against work the client already approved.
Anything the team does the same way twice becomes a skill anyone runs by name. Owners can provision a skill to the whole org from Organization settings, so it’s on for everyone the same day.
The first five: client update, meeting recap, brief review, QA pass and monthly report. Track them in the skill registry (next page).
Recurring work runs on a cadence in Cowork: the weekly client report, the Monday picture, the month-end roundup.
Every schedule has one person who reads its output. A schedule nobody checks is how mistakes reach a client.
Bundle the shared skills into an org plugin and set it to install by default, or mark it required. A new hire gets the agency’s setup on day one instead of collecting it over a month.
Your developers get the same idea in Claude Code: see the Claude Code Cheat Sheet.
| Skill | What it does | Owner | Version | Last checked |
|---|---|---|---|---|
| client-update | Weekly status from the tracker and Slack, in the client’s voice, for the account lead to edit | Account management | v3 | [date] |
| meeting-recap | Decisions, open questions and actions by owner, plus a follow-up draft | Account management | v2 | [date] |
| brief-review | The questions a creative director would ask, and what’s missing or contradictory | Strategy | v1 | [date] |
| qa-pass | Walks a staging link against the approved copy doc and lists what’s wrong | Creative / dev | v2 | [date] |
| monthly-report | Pulls the month’s numbers into the client report template with a first-pass commentary | Ops / finance | v1 | [date] |
Each phase ends when what’s in the “done” column is true, not when the calendar says so. The last column is how it usually goes wrong.
| Phase | What happens | Done looks like | Watch for |
|---|---|---|---|
| Week 0Baseline | Pick three workflows that eat the most hours and time them as they run today: who, how long, how many revision rounds. | Three numbers written down, by the clock. | Guessing instead of timing. Self-reported savings run high. |
| Week 1Foundations | Business seats for everyone, the policy out, five connectors on, a 45-minute kickoff (07). | Everyone signed in on a business seat, and has read the policy. | Tool sprawl: keeping three tools “for now.” |
| Weeks 2 to 4Three pilots | One workflow per team, such as account management, creative and ops. Do it together on live work, then save it as a skill. | Three skills in the registry, each run by someone other than the person who wrote it. | Pilots that never ship because they’re never quite finished. |
| Month 2Make it routine | Put the recurring ones on a schedule. A Project for every active client. Show-and-tell every two weeks. | Two or more scheduled tasks live, each with an owner who reads it. | No owner: schedules and skills nobody checks. |
| Month 3The first role | Re-time the three workflows. Retire the skills nobody runs. Pick one job for an agent to hold end to end, such as client onboarding. | Before and after numbers on three workflows, and a pruned registry. | A policy nobody reads, and nobody re-times anything. |
If you bill by the hour, the time you save comes straight off the invoice. Price the deliverables that got faster by the job, and the margin stays with you.
People will wonder whether this is about doing the same work with fewer of them. Say what’s true for your agency, plainly, in the kickoff. If the plan is to take on more clients with the team you have, say that, and then show it: put the hours that come back into work people want more of. If you don’t know yet, say that too. People can tell.
Day one: a business seat, the plugin installed, the policy read, and three shared skills run with their champion. Nobody should learn the setup by asking around.
The tables download as CSV files that open in Google Sheets or Excel, and Copy pastes them straight into a sheet’s columns. The documents come as Markdown you can paste into a Google Doc or Notion.
The skills, schedules and numbers you built here are what an agent stands on. The next step is one agent that holds a role end to end, such as onboarding a new client or keeping the lead pipeline moving, and comes back to a person where judgment counts. It’s the third way teams use AI on find your starting point, and it goes far better when the team already works from one setup, because the agent inherits the same Projects, connectors and skills everyone else uses.
SourcesClaude Help Center, connectors, organization skills, organization plugins, scheduled tasks and usage analytics · Anthropic Privacy Center, Is my data used for model training? · OPC, Guidelines for processing personal data across borders (2009) · Quebec, Act respecting the protection of personal information in the private sector, ss. 3.1, 12.1, 17 · Canada’s Anti-Spam Legislation, s. 6 · LEGISinfo, Bill C-36.
Thirty minutes, free. We’ll look at how your team uses AI today, pick the three workflows to align first and sketch your 90-day rollout. If it’s not a fit, you’ll still leave with the plan.