The Centaur Model
Pair human taste and judgment with AI speed and automation
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 7
- Confidence
- 98%
The Centaur Model divides work according to comparative advantage. AI handles repetitive, arduous, and time-consuming operations, while a skilled human sets direction, exercises taste, reviews anomalies, and makes consequential decisions. The resulting loop is neither human-only nor autonomous AI: automation prepares, checks, transforms, or accelerates material, and the expert decides what should be accepted or changed. Successful implementation starts with a workflow map and targets tasks whose correctness can be inspected. It preserves human control where context, ethics, originality, and audience judgment matter. The time saved is deliberately reinvested in higher-value work rather than merely increasing volume. Domain knowledge remains essential because an expert must recognize subtle errors and design an appropriate workflow. Without that human layer, the likely output is scalable but undifferentiated AI-generated material.
Origin
Extracted from Masters in Business through O'Shaughnessy's publishing and film examples, where AI accelerates proofreading, consistency checks, plagiarism detection, color matching, and editing while humans retain creative control.
Core principles
- 01Automate arduous work before replacing creative judgment
- 02Keep humans responsible for taste and consequential choices
- 03Use AI to surface issues rather than silently rewrite intent
- 04Reinvest saved time in higher-value creative work
- 05Build domain expertise alongside tool fluency
How to run it
- 1
Map the workflow
List the stages of a recurring process and identify where time is spent, errors arise, and judgment is required.
Pro tip Ask practitioners which tasks they find both tedious and objectively checkable.
Watch out Do not automate an undocumented process that nobody understands end to end.
- 2
Separate labor from judgment
Classify tasks as repetitive transformation, anomaly detection, expert interpretation, or consequential approval.
Pro tip Begin with repetitive work whose output can be compared against a known standard.
Watch out Apparent repetition may still conceal important contextual decisions.
- 3
Establish a quality baseline
Measure current turnaround time, error rate, and output quality before introducing AI.
Pro tip Retain representative examples for side-by-side evaluation.
Watch out Without a baseline, faster output can be mistaken for genuine improvement.
- 4
Automate a bounded task
Deploy AI for one narrow operation, such as detecting typos, flagging contradictions, or matching color across footage.
Pro tip Require structured output that makes review easy.
Watch out Do not let the system silently modify an author's or creator's intended meaning.
- 5
Insert expert review
Have a domain expert validate suggestions, resolve ambiguity, and approve consequential changes.
Pro tip Show the evidence or source context behind every flagged issue.
Watch out A superficial human rubber stamp is not meaningful oversight.
- 6
Reinvest the saved time
Use recovered capacity for research, refinement, storytelling, taste, and other work where humans create differentiated value.
Pro tip Track whether creative quality improves, not only whether output volume rises.
Watch out Using automation solely to flood the market can produce low-value slop.
- 7
Expand carefully
After the bounded task proves reliable, integrate adjacent automations while preserving explicit ownership and review points.
Pro tip Reassess risks whenever the system moves closer to final decisions.
Watch out Success on one task does not establish reliability on a more subjective one.
In the wild
Infinite Books uses AI to correct typos, render a manuscript in galley format, flag contradictions between chapters, and identify possible plagiarism. The system does not resolve substantive incongruities itself; it returns them to the author for a decision.
→ Authors receive review-ready material much faster while retaining authority over their work.
O'Shaughnessy described using AI for laborious post-production tasks such as matching color across footage shot on different days. Human editors remain in the workflow but gain more time for creative decisions instead of mechanical adjustments.
→ Editors complete routine operations faster and devote more effort to artistic quality.
Common mistakes
Automating final judgment
Allowing AI to make unreviewed creative or consequential decisions removes the expertise that gives the combined system its advantage.
Generating undifferentiated volume
Producing more material without human taste creates scalable mediocrity rather than meaningful value.
Using tools without domain knowledge
Tool fluency alone does not equip a person to identify subtle errors or design a trustworthy professional workflow.
Is it for you?
Best for
It is best for workflows containing repetitive analysis or production tasks alongside important creative and domain-specific decisions.
Not ideal for
It is not ideal for fully autonomous use in high-stakes decisions where errors cannot be reviewed or reversed.
From the transcript
“It's going to be AI plus human being. That's where all the great stuff is going to come.”
“you can create a huge edge by having good taste and curating itll”
“AI is a rocket ship for your mind.”
From the episode
A Timeless Collection of Infinite Wisdom with Jim O'Shaughnessy