The third time you run the same playbook, you'll notice something.
You correct the same two things in the output, because the method is right in general and slightly wrong for your company. A month later you run it again, and you make the same two corrections. Because you never wrote them down.
That's the ceiling on borrowed methods, and this lesson is about what's past it.
So far, the method has come from Effy's HR skills library. The policy interview in Basic setup, identity and your AI policy. The workspace builder in How to organize context files. That's deliberate. The fastest way to automate an HR process is to not write the automation yourself. Ask, answer the questions, review the output.
The rest of the course keeps that promise. The process modules ahead lean on a ready playbook for every workflow they teach: reviews, hiring, onboarding, surveys, helpdesk.
But today you make the methods yours. Two moves: turning corrected playbooks into your own skills, and putting them on a clock with your own scheduled tasks. Then a map that shows where all of this is going.
Part 1: Make the method yours, with skills
A skill is a method you never re-explain.
The signal that you need one is unmissable. You're typing the same corrections again, or copying the same instructions into a second project. Anthropic names exactly these two signals as the moment a prompt should graduate into a skill.
You don't write your first skill. You save it.
This is where the handoff from How to organize context files lands. Effy's playbooks carry the expertise: how a 1-1 prep is structured, what a debrief captures, what a policy has to cover. Your corrections carry the specifics: your scale, your cadence, your language. Run the playbook, fix the output, and then, at the end of any task that went well:
"Save what we just did as a skill. Name it, describe when you should use it, and capture the steps, including my corrections."
The corrections are the valuable part. A skill saved from a real run already knows the things you fixed. A skill written from imagination knows nothing.
What's inside one. Three parts, worth knowing when you edit:
→ Name and description. How Claude knows when to fire it. Write the description like a trigger: "use when I ask to turn a finished review into a development plan."
→ The steps. Numbered, ordered, with the why on rules that matter.
→ The standards. Format, tone, what good looks like. One sample output beats a paragraph of description.
What to take, and what to build. 1-1 prep and debrief you already have: they're Effy playbooks, and they're good as they are until your corrections say otherwise. The ones worth making your own first are the company-specific ones: /review-summary and /pdp against your framework, /survey, /handbook-update, /offer-comms.
Don't build all five this week. Save the one matching the task you do most, and use it three times before you trust it.
Part 2: Put it on a clock, with scheduled tasks
A scheduled task runs a job on a schedule, in the cloud, laptop closed, and delivers finished work.
The split that keeps it maintainable: the task holds the when and what, meaning schedule, scope, and where to deliver. The how lives in a skill the task calls.
Method buried inside a task prompt can't be reused, and you'll drift into maintaining two versions of the same procedure.
Five that earn their keep:
→ Monday people brief (weekly). Last week's people activity, joiners and leavers, unanswered employee questions.
→ Review-cycle chaser (weekly, in cycle). Managers who still owe reviews, nudge drafted for your check.
→ Headcount snapshot (monthly). Joiner and leaver counts by team from the HRIS export.
→ Pulse digest (monthly). Responses sorted into three themes, change vs. last month, one action each.→ Policy-renewal watch (monthly). Policies and certifications due in the next 30 days.
Start with the Monday brief. Lowest risk, and every week it lands is proof the system works.
If you set up the weekly context review at the end of How to organize context files, you've already built one of these. Same mechanics, pointed at your workspace instead of your people.
And here's the rule from your AI policy, exactly where it bites: anything touching sensitive data or sending outward stops at a draft. Automate the prep, not the decision.
Part 3: The ladder, or the steps of AI adoption
Everything in this course fits one picture: a ladder with five steps, from organizations where AI is locked behind approvals to ones where most work starts itself and humans steer by intent.
Step 0, Gated. You're waiting. AI is blocked or chat-only, nothing touches real HR data, and whatever gets produced lives on someone's personal laptop. The bottleneck isn't capability. It's approval processes and fear.
Step 1, Assisted. You and Claude, one task at a time, and you watch and review everything. The unlock is real: an afternoon task done between meetings. The bottleneck is your attention, because you read every output.
Step 2, Parallel. You're the orchestrator. Several Cowork tasks run across your project rooms, skills keep the quality consistent, and you review finished drafts, not keystrokes. A week of cycle prep becomes one afternoon. The new bottleneck: reviewing multiple streams of output.
Step 3, Supervised autonomy. You're the manager. Scheduled tasks start work without you, briefs and chasers and digests arrive on their own, and you handle the exceptions. Maintenance work runs continuously in the background. The bottleneck is trust in the loop, and your own decision speed.
Step 4, AI-native. You steer by intent. Most work is kicked off by the system, not by a person, and you set direction and monitor by exception. A quarter-long program becomes kickoff plus check-ins. The bottleneck is guardrails per type of work, at scale.
Three things this ladder tells you.
Where you are. Most HR professionals who "use AI" are at step 1. And most who tried it once and quit judged the whole ladder by step 1's bottleneck. Nobody enjoys reading every output. The point was never to stay there.
Where this course takes you. The three lessons before this one built the step 0 to 1 move properly: setup, identity, data rules, context. Today's two moves, skills and scheduled tasks, are the climb from 1 to 2 to 3. An HR team of one, at step 3, runs an operation that used to take a department's worth of chasing.
The warning built into the ladder. More AI alone never moves you up a step. Each level is unlocked by breaking the next bottleneck and building the next guardrail.
The HR translation: buying a better plan won't get you past step 1. A verification habit (review the draft), a quality bar (skills with standards baked in), and rules that enforce themselves (the AI policy you wrote) will.
Step 4 is the fully AI-native HR function, where the system spots the expiring policy and kicks off its own remediation. It's real, but it isn't a personal setup anymore. It's shared context, team permissions, audit trails, an operation.
When you feel step 3 getting tight, you've outgrown this course. That's the good problem.
Module 1 is done: the system exists. Claude knows who you are, what the rules are, and how your company works, and one review already runs on a clock. What the system hasn't done yet is carry a real process end to end. That's Module 2: one new hire, from signed offer to day 90, run through everything you just built.