01
The win: your scored map
Inventory last week as steps and score each one in the first half hour. The map usually surprises people.
Done looks like Your week, scored and visible.
Workshop
Audit your workflow before you automate it.
Most AI adoption fails because people automate before they audit. Map your real week, score every step, build one automation that earns its place — and write down what deliberately stays human.
The method
Score first. Bake the context once. Then automate one thing well.
The audit lens is simple: for every step in your week, score frequency × pain × AI-fit. High scores earn automation; high-pain-low-fit steps get named and left alone. Then bake your context into a brief a fresh AI session can start from — so you stop re-explaining your project in every chat — and build one automation from it.
The receipt: this is how I run my own week. The skills are public at github.com/iruhdam1/skills, the harness method behind them at madhurimaram.com/harness, and you'll see the actual project starters I kick off real work with.
Agenda
Six blocks, ending in a working automation.
You work on your own week — bring your calendar, not a hypothetical.
02
Read the map
Spot the false positives — high pain, low AI-fit — and pick the one step that actually earns automation. Defend the pick in two sentences.
Done looks like A ranked map and one justified pick.
03
Bake your brief
Context once: problem, constraints, decided things, anti-goals. Built against real project-starter examples, so you see what a brief that holds looks like.
Done looks like A brief a cold AI session starts correctly from.
04
Build the automation
Pick the smallest real shape — a skill, a routine, or a loop — and build it for your top-scored step, kicked off from your baked brief.
Done looks like It runs, on a real case.
05
What stays human
Write the not-automated list with reasons. The judgment boundary is half the deliverable — it's what keeps the automation trustworthy.
Done looks like A defended list of what you didn't automate.
06
The 30-day plan
Adoption beats installation: trigger, owner, review date. Test the automation on one more real case before you leave.
Done looks like A dated adoption note.
You leave with
- A scored map of your real workflow
- A baked project brief that kicks off any AI session correctly
- One working automation, tested on real cases
- A written not-automated list and a 30-day adoption note
For teams
The org variant: a half-day, run as a team.
Workflow inventory across roles, a shared scoring wall, two automations built by pairs, and a not-automated register the team signs off. Deliverables become org property. Priced per engagement — message to scope it.
Fit
Stacks well with
The Product Context Pipeline — when the audit surfaces context-gathering as the bottleneck. Product Lab — the org's playground for what the audit surfaces. This is the natural first workshop for a team.
FAQ
Quick answers
What should I bring?
Last week's calendar and one recurring task that annoys you. That's the raw material.
What tools do I need?
Claude (Pro or better). If you already use Claude Code, the automation block goes further — but it's not required. Free 101 pre-work covers setup.
Will you tell me to automate everything?
No — usually the opposite. The written list of what stays human is a required deliverable, not a footnote.
Is there a recording?
Yes — every session is recorded. Live time is for critique and unblocking.
Seats
Cohorts forming now.
Message with "Workflow Audit" and your timezone — or "org audit" to scope the team half-day.