## Core Definition
Agency is significantly more powerful and significantly more scarce than intelligence. In the AI era, this gap widens: intelligence becomes commoditized ("on tap" via AI), while agency remains the scarce differentiator.
**Source**: Andrej Karpathy, X post, February 24, 2025
## The Thesis
> "I had this intuitively wrong for decades, I think due to a pervasive cultural veneration of intelligence, various entertainment/media, obsession with IQ etc. **Agency is significantly more powerful and significantly more scarce.**"
Karpathy admits he—like most people—overvalued intelligence due to cultural conditioning.
## Why Agency > Intelligence
| Intelligence | Agency |
|--------------|--------|
| Can be measured (IQ, tests) | Harder to quantify |
| Culturally venerated | Often invisible |
| Now commoditized by AI | Remains scarce |
| Answers "can you figure it out?" | Answers "will you act on it?" |
| Prerequisite for competence | Prerequisite for impact |
**The key insight**: Having the answer (intelligence) matters less than acting on it (agency). AI can now provide answers; it cannot provide will.
## The AI-Era Amplification
Garry Tan's addition:
> "Intelligence is on tap now so agency is even more important"
**Implication**: When everyone has access to AI-powered intelligence:
- Intelligence becomes table stakes
- Agency becomes the differentiator
- Those who act decisively with AI outperform those who merely access it
This reframes the [[AI Development Mastery Stack]] question: It's not "can you use AI tools?" but "will you direct them toward meaningful outcomes?"
## Karpathy's Challenge Questions
1. **Are you hiring for agency?** (not just intelligence/credentials)
2. **Are we educating for agency?** (not just knowledge transfer)
3. **Are you acting as if you had 10X agency?** (personal challenge)
## Connection to Existing Frameworks
This thesis provides the **why** behind existing agency frameworks:
- [[Agency as Iteration Without Permission]] — The mechanism (iteration without external prompting)
- [[High Agency Tricycle Model]] — The components (clear thinking, bias to action, disagreeability)
- [[Five Low Agency Traps]] — The blockers (vague, attachment, rumination, social norms, rebellion)
- [[You Can Just Do Things]] — The permission (agency doesn't require approval)
**Agency Over Intelligence** adds: The relative value proposition—agency is the scarcer, more powerful resource.
## Practical Implications
### For Hiring
Stop filtering primarily for intelligence proxies (degrees, test scores). Screen for:
- History of self-initiated projects
- Evidence of acting without permission
- Comfort with ambiguity and iteration
### For Education
Shift from knowledge transfer to agency development:
- Assign open-ended problems, not just solvable ones
- Reward initiative, not just correctness
- Build tolerance for failure and iteration
### For Self-Development
Prioritize agency-building over intelligence-building:
- Act on insights immediately rather than accumulating more
- Practice deciding with incomplete information
- Build the "will to follow through" muscle
## Cross-References
- **Source Document**: [[Agency > Intelligence]]
- **Related Concepts**: [[Agency as Iteration Without Permission]], [[High Agency Tricycle Model]]
- **AI Context**: [[AI Development Mastery Stack]], [[Vibe Coding vs Mastery]]
- **Author**: [[Andrej Karpathy]]
---
*Atomic concept extracted December 2025 from Karpathy's "Agency > Intelligence" post*