AI Leadership Lab

Build the report. Check the evidence. Make the decision.

A hands-on workshop for CEOs, CTOs, founders and department leaders who want to build a useful reporting MVP themselves. Everything happens in the Claude App: chat, Projects and Skills on Day 1, then Cowork to build a narrow agent and a scheduled workflow on Day 2. No prior coding experience is assumed; you never open a terminal. Claude runs the calculator for you, and you review each action before allowing it.

Use a fictional company, Northstar Services, to connect business KPIs, system health and business planning. Approved organisational data is optional. Each block below opens a full tutorial with numbered steps, copyable prompts, saved outputs, verification and troubleshooting. Use your facilitator’s workshop code, add your name and email, and verify your email to unlock the tutorials and practice materials. Your progress saves to your account across devices. Returning learners can sign in with their registered email.

Start here

Download the practice pack (ZIP). Unzip it, read PREWORK.md and keep an untouched copy. The pack contains two reporting periods, source definitions, templates, a starter Skill and plugin, a deterministic calculator and failure exercises.

The answers pack (ZIP) holds the answer key, example outputs and completed Skill and plugin. Open it only after your first attempt and keep it outside your practice folder, so Claude cannot read the answers while you work.

Complete prework 48 hours before the workshop: sign in to Claude Desktop, turn on Settings → Capabilities → Code execution and file creation, confirm Projects, Skills, Cowork, plugin upload and scheduled tasks are available, and choose a recurring decision. Confirm approved data and any read-only connector with your organisation. Use fictional local files if no connector is approved; record live integration as not completed.

Day 1 - Claude App: from decision to reusable method

Day 2 - Claude App in Cowork: from method to tested MVP

The agenda describes teaching time, not a booked event. Cohort invitations confirm dates, timezone, venue and facilitators. Short explanations alternate with practice; approximately 80% of the time is hands-on.

What success looks like

You can trace a report claim to its source, reproduce its arithmetic, show a real Skill/agent invocation, demonstrate an unchanged rerun and inspect an actual scheduled execution. You leave with a business initiative plan and three pilots with owners, measurements, cost limits and stop criteria. Reports stay drafts until reviewed by a person.

The required scheduled task uses your local practice folder, so it runs on your computer: Claude Desktop must be open and the computer awake. A saved schedule is not evidence that it ran. Remote scheduled tasks that read from connectors are an extension for laptop-independent work and require separate source-access and cost decisions. Claude subscriptions and usage are separate from workshop enrolment; availability depends on plan and admin settings. No automated report delivery, production system changes or payment integration is included in the practice workflow.

Source mapping and further reading

The Udemy source contains 17 lectures: 16 relevant topics and one bonus-links lecture excluded from teaching. These original exercises adapt the strategic themes and the supplied 13-module framework; they do not reproduce course videos or transcripts. View the full lecture mapping, also included in the ZIP.

• AI Leader: Generative AI & Agentic AI for Leaders & Founders - Udemy
• Proposed two-day framework
• Custom Skills, Cowork, plugins and sub-agents and scheduled tasks

Source outline checked 12 October 2026; official Claude App documentation checked 14 October 2026. Facilitators recheck interfaces and feature access before each cohort. Completion is formative learning and participation, not an accredited qualification or a production-readiness certificate.