── ── Mental model
Incentive Design
Behavior follows incentives more reliably than character, intent, or training. Get incentives right and mediocre operators produce excellent results; get them wrong and talented teams produce dysfunction. This is Munger's 'Reward and Punishment Superresponse Tendency' — the first and, in his account, most underestimated of his psychological tendencies: never think about anything else before thinking about incentives.
Run Incentive Design on a real problem
Bring something you're actually deciding — free, in the browser.
How it works
Step 1 — Goal and actors: desired outcome · required behavior · actors · time horizon. Step 2 — Map current incentives: rewards (financial, status, autonomy) · penalties · timing · observability. Step 3 — Diagnose alignment gap: what behavior do current incentives rationally produce? where's the mismatch (metric, magnitude, timing)? Step 4 — Design new structure (7-item checklist): (1) alignment (2) measurability (3) timing (4) threshold structure (5) anti-gaming predictions (6) long-short balance (7) tampering defense. Step 5 — Anticipate Goodhart's Law: whatever you incentivize will be optimized — map the most-likely gaming pattern and close it. Step 6 — Implement and monitor: pilot first · monitor 3-6 months · review cycle every 6-12 months · build in actor feedback.
When to use it
- user asks why a team keeps doing the wrong thing despite training
- user is designing compensation, bonuses, or commissions
- user says 'people are gaming the metric' or 'our OKRs aren't working'
- user wants to fix a performance management system
- user asks what incentives are driving a behavior
- user is drafting contracts or platform rules to shape behavior
When not to use it
When the decision is routine and reversible, applying a formal method costs more than it returns.
Worked example
Incentive Design in the 2024–2026 AI Economy
The generative-AI boom of 2024–2026 turned incentive design into a live, high-stakes problem across three surfaces at once: how you reward a model (reward design / RLHF), how you pay scarce AI talent, and how you price the product so customer and vendor pull in the same direction. Each surface is a textbook case of Munger's "Reward and Punishment Superresponse Tendency" — and each has already produced its own flavor of reward hacking, the machine-learning name for Goodhart's Law. This walks all three through the…
Install this skill (free, MIT)
npx skills add deciqAI/knowledge-skillsUseful? Star the repo — stars help other builders find it.
FAQ
Why do incentives beat intentions?
Because incentives operate continuously and unconsciously while intentions require ongoing willpower. People gradually rationalize whatever the reward structure pays for — Munger's point is that the distortion feels like honest reasoning from the inside, which is why good people produce bad outcomes under bad incentives.
What is the most common incentive-design mistake?
Rewarding a proxy metric and expecting the underlying goal: paying for calls made instead of problems solved, lines shipped instead of outcomes, quarterly numbers instead of durable value. Goodhart's Law guarantees the proxy gets optimized at the expense of what it once measured.
How should a founder design incentives?
Start from the behavior you actually need, then ask what a rational self-interested person would do under your proposed structure — including the perverse moves. Test it against gaming before launch, keep metrics few, align payout horizons with value horizons, and audit what the structure is actually producing quarterly.
Related mental models
Intrinsic motivation is a portfolio of distinct sources — not a binary switch.
When you say "I know I should but I just don't do it," the standard explanation is a failure of willpower.
A landing page converts when a visitor, in seconds, understands what it is, who it's for, why it's better, and what to do next — with…
Lateral thinking is the discipline of solving problems by approaching them from unexpected angles — not by going deeper along an existing logical path but by…