Thinking Cheat Sheet

Judgment scales; AI commoditizes execution.

Mental Models

Fundamentals

Model	Core Idea
First principles	Strip away assumptions; build from bedrock truths
Inversion	Instead of “how do I succeed?” ask “how would I fail?”
Second-order effects	Then what? And then what after that?
Opportunity cost	Choosing X means not choosing Y
Reversibility	Two-way doors (cheap to undo) vs one-way doors (commit carefully)
Leverage	Small inputs, large outputs — where’s the fulcrum?

Technical

Model	Core Idea
CAP theorem	Distributed systems: pick two of consistency, availability, partition tolerance
Amdahl’s law	Speedup limited by the part you can’t parallelize
Premature optimization	Measure first; optimize the bottleneck
Leaky abstractions	All abstractions fail; know what’s underneath
Worse is better	Simple, working beats perfect, unshipped
Gall’s law	Complex systems that work evolved from simple systems that worked

Decision-Making

Model	Core Idea
Regret minimization	Which choice will 80-year-old you regret less?
Eisenhower matrix	Urgent vs important — different axes
Satisficing	”Good enough” beats endless optimization
Reversible defaults	When uncertain, choose what’s easiest to undo

Compound vs Linear Value

Work that keeps paying dividends vs. work that’s consumed once.

Linear Value	Compound Value
Shipping a feature	Establishing a pattern others follow
Fixing a bug	Creating monitoring that catches bugs early
Writing code	Writing documentation that trains 10 engineers
Solving today’s problem	Preventing tomorrow’s category of problems
Answering a question	Writing docs so 50 people don’t ask it
Helping one person	Building tools that help everyone

The test: When you leave, does your impact persist? If it requires your ongoing presence to maintain value, it’s linear. If it continues without you, it compounds.

Levels of help:

1. 1:1 — Answer a colleague’s question
2. 1:many — Write documentation so 50 people don’t ask
3. Systemic — Build tools that prevent the question from arising

How to shift: Each time you solve a problem, ask: “Can I create something that solves this for everyone?” The answer isn’t always yes—but asking changes what you build.

Systems Thinking

Core Concepts

Feedback loops — Outputs become inputs

• Reinforcing: growth spirals (or death spirals)
• Balancing: self-correcting toward equilibrium

Stocks and flows — Accumulations and rates of change

• Don’t confuse the bathtub (stock) with the faucet (flow)

Delays — Effects lag behind causes

• Systems overshoot because feedback arrives late

Emergent behavior — Whole > sum of parts

• Local optimization ≠ global optimization

Questions to Ask

• Where are the feedback loops?
• What’s the delay between action and effect?
• Who are the stakeholders I’m not seeing?
• What happens if this succeeds beyond expectations?
• What’s the system optimizing for? Is that what we want?

Traps

Trap	Description
Local optimization	Improving your part while harming the whole
Metric fixation	Goodhart’s law: measure becomes target, ceases to be useful
Linear thinking	Assuming proportional cause/effect in nonlinear systems
Ignoring delays	Impatience leads to over-correction

Asking Good Questions

Different questions serve different purposes. Choose deliberately.

Question Types

Type	Purpose	Example
Clarifying	Surface hidden assumptions	”What would have to be true for this to work?”
Reframing	Shift the problem itself	”Are we solving the right problem?”
Aligning	Build consensus	”What concerns need addressing to move forward?”
Unlocking	Develop others	”What do you think should be done?”
Preventing	Stop expensive mistakes early	”What would we see if this was failing?”
Diagnostic	Isolate root cause	”What changed? Can I reproduce it?”

The expert knows which questions to ask, not just how to answer.

Before Building

• What problem are we actually solving?
• Who has this problem? How do they cope today?
• What would “done” look like?
• How will we know if it worked?
• What’s the simplest thing that could possibly work?

Before Deciding

• What would have to be true for this to be the right choice?
• What are we optimizing for? What are we sacrificing?
• Is this a one-way or two-way door?
• What do we believe that might be wrong?
• Who disagrees? Why?

When Stuck

• What do I actually know vs assume?
• What question am I not asking?
• What would I try if I couldn’t fail?
• What would a beginner do?
• What would I advise a friend in this situation?

Debugging Questions

• What changed?
• What did I expect? What happened instead?
• Can I reproduce it?
• What’s the smallest case that shows the bug?
• Am I solving the symptom or the cause?

Minding the Real Goal

The Goal Stack

Immediate task    → "Deploy the feature"
Project goal      → "Increase user engagement"
Business goal     → "Grow revenue"
Actual goal       → "Build something people want"
Life goal         → "Do meaningful work"

Working on the wrong level wastes effort. Zoom out regularly.

Traps

Trap	Example
Task fixation	Finishing the ticket vs solving the user’s problem
Vanity metrics	Optimizing for lines of code, story points, commits
Sunk cost	Continuing because you’ve already invested, not because it’s right
Cargo culting	Copying practices without understanding why they work
Resume-driven development	Choosing tech for your CV, not the problem

Reframes

• “What would success look like if this feature didn’t exist?”
• “If we couldn’t build this, how else might we achieve the goal?”
• “What’s the user actually trying to accomplish?”
• “Will this matter in a week? A year?”

Learning with AI

The New Division of Labor

AI handles well	Humans handle better
Syntax, boilerplate	Architecture, design
Lookup, recall	Judgment, taste
First draft	Final edit
”How to X in Y"	"Should we X at all?”
Speed	Direction

Skills That Compound

• Verification — Can you tell if the output is correct?
• Decomposition — Breaking problems into AI-sized chunks
• Iteration — Refining output through dialogue
• Integration — Combining AI output into coherent systems
• Knowing when to go deep — Some things need human understanding

Effective AI Collaboration

Do:

• State context and constraints upfront
• Ask for reasoning, not just answers
• Verify against your mental model
• Iterate — first answer is rarely best
• Use AI to explore options, then decide yourself

Don’t:

• Trust without verification
• Outsource understanding
• Skip building mental models
• Accept complexity you can’t maintain
• Let AI make one-way-door decisions

The Paradox

You must know enough to verify AI output, but AI reduces the need to memorize.

Resolution: Learn principles deeply, details just-in-time.

Daily Practice

1. Before starting: What’s the real goal here?
2. When stuck: What question am I not asking?
3. After finishing: What would I do differently?
4. Weekly: What did I learn? What’s still fuzzy?

---

“The formulation of a problem is often more essential than its solution.” — Einstein

See Also

• Reasoning — Fallacies, cognitive biases, and debiasing techniques
• Debugging — Apply systematic thinking to find bugs
• Orchestration — Systems thinking applied to agent coordination
• Learning — How learning builds on mental models
• Problem-Solving — Applying thinking frameworks to problems
• Reasoning Lesson Plan — Progressive lessons on clear thinking