Set up ChatGPT Desktop or Claude Desktop for real work, then turn one weekly task into a reusable workflow.
The finish line
You will leave with clear privacy settings, standing instructions, one approved connection, a reusable workflow, and a Chief of Staff prompt.
Progress is saved in this browser.
From chat to agent
One desktop AI home base
Clear rules for private work
Standing context and flexible reasoning
One approved connection to real work
One reusable weekly workflow
A path to a cross-tool Chief of Staff heartbeat
Each step makes the agent more useful while keeping you in control.
The guide
Step 1 of 5 · 5 min
Choose ChatGPT or Claude as the home base for your AI work.
You will move faster when your chats, files, connections, and recurring work live in one place. Pick one to start. You can always add the other later.
Do this
Step 1 of 5 · 5 min
Choose ChatGPT or Claude as the home base for your AI work.
Do this
Remember: Pick the tool your team supports and your data policy allows. The workflow matters more than the logo.
Step 2 of 5 · 7 min
Decide what the agent can remember, retain, and learn from.
More context makes an agent more useful. It also raises the stakes. Set the limits before you connect anything.
Do this
Step 2 · practical defaults
Step 2 of 5 · 7 min
Decide what the agent can remember, retain, and learn from.
Do this
Remember: Never paste passwords, account numbers, private health records, or confidential information you are not authorized to share.
Step 3 of 5 · 10 min
Add your role, audience, voice, and decision rules once.
Good standing instructions save you from repeating yourself. They give the agent enough context to be useful without forcing every task into the same answer style.
Do this
Step 3 of 5 · Run it
Custom Instructions example
About me: I run a small bakery and write our weekly newsletter. Who it is for: Local customers. How to respond: Keep it warm and casual. Use short paragraphs and plain language. One thing to always include: Suggest a seasonal item we can feature.
Adaptive Reasoning Protocol
Use the Adaptive Reasoning Protocol: Assess the request type, then apply the appropriate method: If solving, analyzing, or debugging → Atom of Thought: Decompose into atomic reasoning units. For each atom: State the logical component Validate independence Verify correctness Then synthesize atoms into final answer. If explaining, learning, or teaching → Feynman Loop: Explain as if teaching a curious beginner. For each cycle: Use a concrete analogy Flag confusion points Ask questions that reveal gaps Then compress into a teachable snapshot. If both are needed → Chain them: First solve via Atom of Thought, then explain the solution via Feynman Loop.
Step 4 of 5 · 8 min
Let the agent read one approved tool that holds real work.
Without context, AI waits for you to paste everything. One approved connection lets it find current information and link you straight to the evidence.
Do this
Step 4 of 5 · 8 min
Let the agent read one approved tool that holds real work.
Do this
Remember: Start read-only. Add permissions only when the workflow and review step are clear.
Step 5 of 5 · 12 min
Save the instructions so the agent can do the work the same way next time.
The real gain starts when a good conversation becomes a repeatable way of working. Save the context, expected output, and quality bar so you are not starting over next week.
Do this
Step 5 of 5 · 12 min
Save the instructions so the agent can do the work the same way next time.
Do this
Remember: It is ready when another person can run it and recognize a good result.
Bonus · design the next workflow
It gives the agent room to clarify the job before jumping to an answer. Replace the example with one bottleneck from your week.
How Might We prompt
How might we turn my meeting notes into a clear weekly update? Before you write, ask me up to three questions about the audience, format, and what matters most. Then give me a reusable prompt I can run each week.
The payoff · one clear heartbeat
Do
A decision is waiting on you.
Direct source linkDelegate to AI
The agent has enough context to draft it.
Direct source linkDiscuss
The tradeoff needs a human conversation.
Direct source linkStart read-only. Let the agent gather, sort, and link. Keep approvals, messages, task changes, and other outside actions behind human review.
Copy, customize, run
Quick-start prompt
Turn this thread into a heartbeat, I want you to check my emails, slack, clickup every 3 hours and tell me what to prioritize. Make sure everything has a link
Full Chief of Staff prompt
Loading the full prompt...
Your workspace is ready
Use the same pattern for your next bottleneck: name the job, set the rules, connect only what it needs, save the method, and review the result.
Go deeper
Build confidence with Claude's core controls, projects, files, and reusable workflows.
Open trainingBuild with evidence
Turn one Job Brief into a bounded, tested prototype with a clear human handoff.
Open trainingGet support
Keep learning with practical training, live support, and other people doing the work.
Open training