Think First, Delegate Selectively
Preserve human judgment while delegating low-value production work to AI.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 93%
First classify the work by how much personal judgment, original thinking, and authentic voice it requires. For consequential decisions or sensitive conversations, develop your own point of view before involving AI. Then use the model as a sparring partner: ask it to test assumptions, anticipate reactions, sharpen reasoning, and push back where the idea is weak. Retain responsibility for choosing among its suggestions. For standardized outputs such as recurring business reviews, invert the relationship by delegating most of the drafting and acting primarily as reviewer and verifier. The dividing line is not whether AI can produce words, but whether the production of those words carries more value than the underlying thinking. Across both modes, humans remain accountable for verification and sign-off, while AI supplies leverage, alternatives, and critique.
Origin
Extracted from Lenny's Podcast, where Dianne Penn contrasted using Claude to challenge her point of view in judgment-heavy situations with delegating standardized monthly business-review writing.
Core principles
- 01Form your own point of view when personal judgment matters.
- 02Use AI as a sparring partner rather than an automatic authority.
- 03Ask the model to challenge assumptions and improve the outcome.
- 04Delegate standardized production work more aggressively.
- 05Review and verify delegated outputs before signing off.
- 06Preserve personal voice when the communication represents you.
How to run it
- 1
Classify the work
Determine whether the task's value lies mainly in judgment and personal voice or in standardized production. Consider the consequences of a weak or incorrect output.
- 2
Form an initial view
For high-judgment work, write down your own position, assumptions, and intended outcome before asking the model for help.
- 3
Invite challenge
Ask the model to identify weak assumptions, anticipate reactions, propose alternatives, and explain where it disagrees. Treat it as a thinking partner rather than an approval machine.
- 4
Exercise judgment
Compare the critique with your goals and context, then decide what to adopt. Preserve your tone and responsibility for the final decision.
- 5
Delegate standardized production
For repeatable outputs, provide source material, structure, and quality criteria, then let AI produce the first complete draft.
- 6
Review and verify
Check evidence, calculations, interpretation, omissions, and tone before signing off. Make accountability explicit even when AI produced the text.
In the wild
Penn uses a Claude skill informed by Crucial Conversations to prepare for difficult managerial discussions. She considers her own objective and judgment, then asks Claude to help anticipate reactions, find the right level of detail, build trust faster, and make the message more direct. She does not necessarily use every suggestion.
→ Claude functions as personalized coaching while Penn retains responsibility for the conversation and her authentic management style.
A leader provides the relevant metrics, decisions, and standard reporting format to an AI system. Because the recurring prose is less valuable than the underlying analysis, the model drafts the review end to end. The leader spends time checking evidence, correcting framing, and approving the final document rather than manually producing every sentence.
→ Routine writing consumes less time while human attention remains focused on thinking and verification.
Common mistakes
Letting AI define the initial position
On judgment-heavy work, beginning with the model's conclusion can replace independent thought with anchoring. Establish your own perspective first.
Using AI only for agreement
A useful thinking partner should expose weak assumptions and add new ideas. Ask for pushback rather than reassurance.
Delegating without verification
Standardized writing may be safe to automate, but factual accuracy and organizational accountability still require human review and sign-off.
Is it for you?
Best for
It is best for managers, product leaders, and knowledge workers deciding how much of a task to delegate to an AI system.
Not ideal for
It is not ideal for decisions requiring licensed expertise, confidential human accountability, or verification that the user cannot competently perform.
From the transcript
“I might uh um come up with my own POV first and then work with Claude through that.”
“what you don't want is like a AI that just agrees with you, right?”
“I'm more of a reviewer and a verifier of that information.”
From the episode
Anthropic’s first technical PM on token maxing, the jagged edge, and living in the future