You’ll leave with 10× better AI answers, the “Magic Prompt,” and a Chief of Staff agent.
Each gear handles a larger unit of work, from one quick question to a continuous system. Human judgement sets direction and reviews the result at every level.
Match the gear to the problem
What AI Agent Skills Are
An agent is the teammate. A Skill is the playbook.
Agent
Capable teammate
Skill
Reusable playbook
Result
Reliable output
Example Skill
Teach it once. Reuse it. Improve it after every run.
The Biggest Mistake
“I need to make this thing produce an answer.”
“I need to define the outcome, context, boundaries, and review.”
Think Roomba, not remote control.
A Roomba owns the objective: clean the floor. You set the boundaries and review the result instead of deciding which room comes next and which vacuuming pattern to use.
The management model
The leadership model
Know what AI can do, where it breaks, and what is arriving next.
Supply role, goal, constraints, context, taste, and prior attempts.
Choose which plausible answer fits your customer, culture, and moment.
Give the agent only what the work needs
Give the team eyes, hands, and a clearly bounded workspace.
Remember: Start with the sources the workflow already relies on. Add only the actions needed to complete the work. Be explicit about which actions are read-only, reversible, approval-gated, or external.
Calendar, email, docs, analytics, code.
Search, draft, edit, run, verify.
Read, draft, change, publish.
10× your AI answers
Give the agent a repeatable way to reason and explain. Place this in the primary surface’s system-prompt or custom-instructions field, then ask what instructions it is operating under.
Remember: The agent reports the protocol and returns substantively different approaches instead of one linear answer.
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.
The Magic Prompt: Outcome-Oriented AI
Put it together
“How might we rewrite this sales page to increase qualified calls without changing the offer, based on customer interviews, PostHog behavior, and our best-performing page?”
Turn scattered context into clear priorities
Do this
Chief of Staff starter brief
CHIEF OF STAFF AGENT OUTCOME Give me a ranked brief that shows what matters now, what is blocked, and what can wait. CADENCE Run every [morning / workday] at [time]. APPROVED SOURCES - Calendar: [account / calendars] - Email: [account / labels / time window] - Slack or Teams: [workspace / channels / time window] - Tasks: [system / projects / lists] - Active commitments: [source of truth] DECISION RULES DO: I should personally act because: DELEGATE: Another owner should act because: DELETE / DEFER: This can wait or stop because: FOR EVERY ITEM, INCLUDE 1. Do / Delegate / Delete 2. Why it matters now 3. Evidence with a direct source link 4. Blocker or dependency 5. Recommended next step and owner BOUNDARIES - Read source systems only. - Do not send, post, invite, edit, delete, or change records. - Separate facts from recommendations. - Do not turn weak evidence into a commitment. STOP RULE Stop and ask one clear question when a source is unavailable, two sources conflict on a critical fact, ownership is unclear, or the next step would create an external change. DONE The brief is easy to scan. Every priority links to evidence. Conflicts are visible. I can decide the next move without reopening every source.