
I recently appeared on a UK podcast called AI Superwoman. Host Lily Patrascu is a woman who dove right in to ensure her audience understood how to advance with AI. It was a pleasure making this guest appearance, and I think I held my own when Lily hit the hard questions right out of the gate.
We discussed when to use AI as a tool, an Agent, or for automation. Done correctly, this balance will empower users to increase revenue, reduce costs, and streamline workflows. To simplify the process, I shared my three-step decision-making framework to help you determine which approach to use AI based on your circumstances at any given time.

Before the show, Lily asked me to create the Ten Commandments for getting work done with AI. By the time we addressed the topic, we had limited time left on the show, so I wanted to give you the Ten Commandments here. Additionally, the entire show is available below.
1. Start with the outcome.
- Why: AI can’t hit a target you haven’t set.
- How: Define the job-to-be-done, success metric, deadline, and constraints in one sentence.
2. Pick the right level: Think, Do, or Rule.
- Why: Matching tool to task saves time and errors.
- How:
- If you need ideas/clarity → AI.
- If you need actions across apps → AI Agent.
- If it’s fixed steps with clear rules → Automation (VBA/Zapier).
3. Feed it structured context.
- Why: Better inputs = better outputs.
- How:
- Give role
- Goal
- Audience
- Tone
- Examples
- Requested format (e.g., table, checklist)
4. Keep a human in the loop for judgment calls.
- Why: Brand, ethics, and nuance still need you.
- How:
- Set approval points:
- “AI drafts → You review → Then send/post.”
- Set approval points:
5. Trust, but verify.
- Why: AI can sound confident and be wrong.
- How:
- Spot-check facts.
- Re-run math.
- Test on a small sample before scaling.
6. Protect data like it’s cash.
- Why: Leaks, compliance, and client trust are on the line.
- How:
- Remove PII/secrets
- Use least-privilege access
- Log who/what the agent can touch
7. Design for repeatability.
- Why: One-offs don’t scale; systems do.
- How:
- Save prompt templates
- Name variables
- Version your workflows
- Keep an audit log
8. Measure what matters.
- Why: If it doesn’t improve results, it’s a toy.
- How: Track to retire what underperforms.
- Time saved
- Error rate
- Cost per task
- Outcome quality
9. Pilot small, then automate.
- Why: Early wins build confidence and reveal edge cases.
- How: Manual SOP → macro/automation → add agent logic → scale team-wide.
10. Fail safely.
- Why: Mistakes happen—contain them.
- How:
- Set guardrails (allowed actions, rate limits)
- add fallbacks
- add a kill switch
- keep backups
SUMMARY: Define the outcome, match tool to task (Think/Do/Rule), give great context, keep humans and safety in the loop, and scale only what the numbers prove works.
If you’re interested in listening to the entire podcast…
Copyright © 2025 by CJ Powers

