10x Productivity

with Claude & ChatGPT Agents + Loops

Mistake 01

Treating AI like a tool instead of a team.

Tool thinking

“I need to make this thing produce an answer.”

→

Team thinking

“I need to define the outcome, context, boundaries, and review.”

1
Mistake 02

No process for continuous improvement or change management.

One-off use
A real loop
Run
Review
Capture
Update
Repeat
2
The management model

AI teams improve the same way people do.

  • 01Context
    What matters here?
  • 02Direction
    What outcome do we own?
  • 03Feedback
    What should repeat next time?
Case study 01

Chief of Staff

Watches

Calendar, email, Slack, tasks, and active commitments.

Produces

A ranked brief: what matters, what is blocked, what can wait.

Leader move

Teach the decision rules behind “urgent” and “important.”

Case study 02

Change Requester

01Intake
02Clarify
03Scope
04Implement
05Review

Turn a messy request into review-ready work with an explicit approval boundary.

Case study 03

Sales Page Optimizer

Move from “make this better” to evidence-led conversion work.

01 Customer language
02 Behavior data
03 Best-performing copy
Case study 04

Error Fixer

Sees

Error analytics, logs, code, tests, and deployment state.

Does

Reproduce → isolate → patch → validate

Boundary

No attributable, reproducible error? Return NO CHANGE.

Case study 05

Continuous Improvement

Where are you still doing work your AI team could own?

3 steps

Step 01

Set up Connections & Tools

Give the team eyes, hands, and a clearly bounded workspace.

  • 01Connect sources
    Calendar, email, docs, analytics, code.
  • 02Connect actions
    Search, draft, edit, run, verify.
  • 03Set boundaries
    Read, draft, change, publish.
Step 02
“How might we VERB OUTCOME with CONSTRAINTS based on SOURCES”
Action Definition of done Rules + boundaries Evidence
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?”
Step 03

Iterate with GOOD_EXAMPLES

Coach the pattern you want repeated, not just the output in front of you.

  • 01Mark what worked.
  • 02Explain what should change.
  • 03Save the best version.
  • 04Update the reusable skill.
The loop

Every review should improve the next run.

Now

Do this first …

  1. Choose one recurring workflow with a reviewable output.
  2. Connect the minimum sources and tools it needs.
  3. Run one outcome-first prompt.
  4. Save the best result as your first GOOD_EXAMPLE.