Source: Be Datable, building on Andrej Karpathy's "Agency > Intelligence" insight

Capability

Agency vs Intelligence

The Threshold Principle: Why action beats planning, but only after you clear the bar

Why This Matters for Your Career

The people getting promoted aren't the smartest in the room. They're the ones who ship. Every organization has brilliant analysts who never influence decisions, and average performers who drive massive change. The difference is agency.

But there's a catch most people miss: agency without sufficient competence is worse than doing nothing. The diagram above shows this as the "threshold line." Below that line, increasing your bias toward action just creates more mess to clean up. Above it, every increment of agency multiplies your impact.

This applies to AI tools, teams you manage, and most importantly, yourself.

Take the Assessment

Find out where your AI tools fall on the Agency vs Intelligence spectrum.

Start Assessment

The Four Quadrants

Low Intelligence, Low Agency

Where most chatbots lived for five years. "If-then-else" loops pretending to be smart. These destroyed consumer trust.

Career equivalent: The person who doesn't know much and doesn't do much. Safe but forgettable.

High Intelligence, Low Agency

Where most AI models sit today. Impressive knowledge, zero initiative. They wait for your prompt.

Career equivalent: The brilliant analyst who writes great memos that nobody reads. Smart but ineffective.

Low Intelligence, High Agency

The danger zone. Systems (and people) that act confidently without adequate understanding. Creates expensive mistakes.

Career equivalent: The eager new hire who sends emails to the CEO without understanding company politics.

High Intelligence, High Agency

The goal. Systems that think critically AND execute across platforms without constant direction.

Career equivalent: The person who identifies problems, proposes solutions, and implements them. The one everyone wants on their team.

How to Use This Framework

For Your AI Tools

  1. Stop using free models for important work. They're below the intelligence threshold. The $20/month for Claude or GPT-4 is less than your unused Peloton membership.
  2. Audit your tools: Do they just answer questions, or can they take action? Look for API integrations, automation capabilities, tool use.
  3. Add agency incrementally. Start with low-risk tasks (scheduling, drafting) before high-risk ones (customer communication, financial decisions).

For Your Team

  1. Map each person on the quadrant. Where are they?
  2. For those below the threshold: invest in training before expanding their scope.
  3. For those above but low-agency: give explicit permission to act. Many smart people wait for approval that never comes.
  4. For high-agency people: ensure they have adequate information and context before they move.

For Yourself

  1. Honestly assess: Are you above the competence threshold for the decisions you're making? If not, learn first.
  2. If you're competent but not advancing, your problem is probably agency, not intelligence. Ship something this week.
  3. Increase agency in small steps. Don't wait for permission. Act, then ask forgiveness if needed.

Self-Assessment

Answer these questions to evaluate your position.

In the last month, how many ideas did you have vs. how many did you ship?

If the ratio is worse than 10:1, you have an agency problem.

When was the last time you did something at work without being asked?

If you can't remember, you're in the low-agency quadrant.

When you act independently, do outcomes generally improve or get worse?

If worse, you may need to cross the intelligence threshold first.

Does your manager spend more time directing you or cleaning up after you?

If cleaning up, dial back agency and invest in learning.

The Core Principle

Agency consistently outperforms intelligence, but only after crossing a minimum intelligence threshold.

This principle explains why some organizations fail when they give junior employees too much autonomy, and why others stagnate when they don't give senior employees enough.

The threshold isn't fixed. It varies by task complexity, risk level, and reversibility of decisions. Posting on social media has a lower threshold than negotiating a merger. Adjust accordingly.

Agency beats intelligence, but only above the threshold.

Ship more. But make sure you know enough first.

Sources & Further Reading

Get more frameworks like this

Weekly frameworks for becoming more data-able in the age of AI.

Related Frameworks