mouseover

← All challenges

Hands-On Agentic AI for Leaders

Shift from chatting with AI to building production agent workflows you can deploy, govern, and reuse across your org.

Weekly time
~6 hrs/week
For
C-suite, leaders, and senior ICs who learn by building, not by watching.
Prerequisites
A ChatGPT, Claude, or Gemini subscription. No coding required.
Structure
30 days · 4 milestones · 11 checkpoints · 47 steps
Start in mouseover →

Opens the challenge in mouseover. Download mouseover first if you have not yet.

Week 1

Foundations & Workflow Identification

Audit recurring work, identify high-value workflow candidates, set up your AI builder stack.

End-of-week outcome: A ranked list of 3 workflow candidates plus a working builder environment.

Checkpoint 1 · ~2 hrs

Audit your recurring work

You will have: A ranked list of 3 candidate workflows scored on frequency × pain.

Show 4 steps
  1. Step 1 List 10 things you do every week that take more than 20 minutes.
  2. Step 2 Score each on frequency (how often) and pain (how draining or error-prone).
  3. Step 3 Mark the ones you would describe out loud to a colleague the same way every time. These are workflow candidates.
  4. Step 4 Pick the top 3 and write a one-sentence description of each.
Checkpoint 2 · ~3 hrs

Apply the 7-step workflow lifecycle

You will have: Your top candidate decomposed into a written workflow spec.

Show 5 steps
  1. Step 1 Take the top candidate and write out the workflow as a numbered list of steps.
  2. Step 2 Mark each step as "decision" (judgment required) or "mechanical" (rule-following).
  3. Step 3 Identify which mechanical steps an LLM could do with a clear prompt.
  4. Step 4 Identify which decision steps still need human review and what good looks like for each.
  5. Step 5 Write a one-page workflow spec: inputs, steps, outputs, review points.
Checkpoint 3 · ~2 hrs

Set up the builder stack

You will have: A working environment where you can author, version, and test AI workflows.

Show 4 steps
  1. Step 1 Install Cursor or VS Code. Get comfortable opening, saving, and committing files.
  2. Step 2 Create a GitHub account if you do not have one and make a private repo for your AI assets.
  3. Step 3 Pick your primary model: Claude, ChatGPT, or Gemini. Confirm you can access the API or platform of your choice.
  4. Step 4 Run one short prompt as a saved file in your repo. Commit it. You now own that asset.
Week 2

Agent Skills & Deterministic Workflows

Package skills as files you own. Build workflows you can test, version, and hand off.

End-of-week outcome: Two production workflows: one deterministic, one human-in-the-loop.

Checkpoint 1 · ~3 hrs

Build your first agent skill

You will have: A reusable, file-versioned skill that beats raw chat on 5 real inputs.

Show 5 steps
  1. Step 1 Pick a skill type from your workflow: research, summarisation, analysis, or drafting.
  2. Step 2 Write the skill as a prompt-in-a-file with explicit role, inputs, format, and constraints.
  3. Step 3 Run it on 5 real inputs you actually deal with. Capture the outputs.
  4. Step 4 Compare against raw "ask ChatGPT" on the same inputs. Note where the skill version wins.
  5. Step 5 Iterate the prompt once. Commit the new version. You now version a skill.
Checkpoint 2 · ~3 hrs

Build a deterministic workflow

You will have: A 2-to-3-skill chain that runs end-to-end on real input with a pass rate you can quote.

Show 4 steps
  1. Step 1 Chain 2 or 3 of your skills with explicit handoffs (output of A becomes input of B).
  2. Step 2 Define the fixed rules: what runs always, what skips, what triggers a stop.
  3. Step 3 Run on 5 real inputs. Log pass and fail for each step.
  4. Step 4 Write down the pass rate. This is your baseline.
Checkpoint 3 · ~2 hrs

Add a human review checkpoint

You will have: A collaborative workflow where a human approves before the final step ships.

Show 4 steps
  1. Step 1 Identify the one step in your workflow where human judgment matters most.
  2. Step 2 Design the handoff: what does the human see, what do they decide, where do they signal "approved"?
  3. Step 3 Run the collaborative version with a real colleague on a real input.
  4. Step 4 Note what the colleague pushed back on. Update the prompt so the AI catches it next time.
Week 3

Autonomous Systems

Deploy single-agent and multi-agent workflows for your highest-value work.

End-of-week outcome: A deployed single-agent workflow running unattended and a multi-agent design tested in dev.

Checkpoint 1 · ~3 hrs

Deploy a single-agent workflow

You will have: A workflow that runs unattended on a real input and produces a real output.

Show 4 steps
  1. Step 1 Pick the platform that fits your use case: Claude Projects, custom GPT, M365 Copilot, or Gemini.
  2. Step 2 Deploy your workflow with all the skills and rules from week 2.
  3. Step 3 Trigger it on a real input. Measure how long it took, what it produced, where it diverged.
  4. Step 4 Document the trigger, expected inputs, and expected outputs.
Checkpoint 2 · ~3 hrs

Design a multi-agent system

You will have: A documented 3-agent handoff with a tested failure recovery path.

Show 5 steps
  1. Step 1 Pick a workflow too big for one agent. Often: research → synthesis → output.
  2. Step 2 Design 3 agents with explicit roles, handoffs, and inputs/outputs.
  3. Step 3 Implement the handoffs. Could be as simple as file passes or as rich as a tool-call chain.
  4. Step 4 Run on one real input. Inject a known failure midway. Document the recovery path.
  5. Step 5 Decide which workflows are worth this complexity and which are not.
Week 4

AI Registry & Governance

Make your skills, prompts, and agents discoverable, reusable team assets.

End-of-week outcome: A living AI Registry in Notion with a 30-day team rollout plan.

Checkpoint 1 · ~3 hrs

Build the AI Registry

You will have: A Notion database listing all your skills, prompts, and agents with status, owner, and use.

Show 4 steps
  1. Step 1 Create a Notion database with columns: name, type (skill/prompt/agent), status, owner, use case, link to file.
  2. Step 2 Import every asset you built in weeks 1 through 3.
  3. Step 3 Tag each by maturity: draft, tested, production.
  4. Step 4 Add a "last reviewed" date column so quality stays current.
Checkpoint 2 · ~2 hrs

Make it discoverable

You will have: A team-facing landing page on the registry with 3 example usages.

Show 4 steps
  1. Step 1 Write a one-page intro to the registry: what is here, when to use it, who owns it.
  2. Step 2 Add 3 example usages with the exact inputs and outputs.
  3. Step 3 Share with 2 colleagues. Ask them to find a specific asset. Note what they fail to find.
  4. Step 4 Fix the discoverability gaps you observed.
Checkpoint 3 · ~2 hrs

Plan the team rollout

You will have: A 30-day rollout plan with 3 candidate workflows and first office hours scheduled.

Show 4 steps
  1. Step 1 Pick 3 workflows from your registry that are ready for team adoption.
  2. Step 2 For each, write: target team, value sentence, expected adoption blockers.
  3. Step 3 Draft a 30-day rollout plan: week 1 announce, week 2 train, week 3 use, week 4 retro.
  4. Step 4 Schedule your first office hours session and post it.

Ready to start?

You have read the plan. The next thing to do is open the first checkpoint.

Start in mouseover →

Opens the challenge in mouseover. Download mouseover first if you have not yet.

Browse other 30-day challenges ← Back to all challenges