The Causally Culture Handbook

Preface: this is Causally’s internal culture handbook (v1.0), lightly edited for public reading — team blueprint and talent pipeline stripped, the cultural and rhythmic core kept whole. Background in the earlier post, Why Causally.

On the Name

Causally — the adverb, “in a manner that follows cause and logic” — is the attitude behind everything we do. The Chinese name, 溯理 (Sùlǐ), means the same thing: trace the cause to its root, and act in accordance with it.

Mission and Operating Model

I laid out the two missions fully in Why Causally. Here’s the one-line compression:

Five people co-evolving with AI, producing what a hundred-person traditional team can’t. And entering vertical after vertical with AI-native product capability.

The operating model deserves a few more lines. Causally isn’t just a company; it’s an experiment in organizational form and intelligent collaboration:

AI is not just a tool. It is the new productive force itself, replacing what was previously the key input — human cognitive labor.
When AI takes over the thinking and execution of many tasks, the human role upgrades into management: no longer “doing tasks,” but defining goals, allocating and managing, ensuring quality, and taking responsibility.
Humans handle intent, judgment, learning, and creation. AI handles execution, scaling, teaching, and codification.
Both business and life are infinite games — and at their core, reinforcement learning: direction, feedback loops, iteration speed. The organization as a whole evolves through real-world feedback and continuous iteration.

At Causally, every person is the responsible party for a domain — a Domain Owner: understands the goal, can decompose the problem, can drive AI to deliver the result, and stays accountable for quality all the way through. Like the head of a department — except the team underneath is now a scalable, replicable, continuously-growing army of AIs.

The Standards We Hold Ourselves To

Everyone’s starting point is T-shaped: at least one real specialty, plus a wide horizon. With AI’s help, you deepen in adjacent breadth over time, becoming “comb-shaped.” Across the team, the outer ranges overlap; the cores (the thinking inside each person) are diverse and complementary.

We use these five dimensions to look at ourselves, and at anyone joining:

Dimension	Definition	Behavioral signature
Outcome Thinking	Start from the “goal value” / “problem to be solved,” not a task list	Clarify the problem and the bar first, then pick the optimal path. Once the goal is clear, the stages decompose themselves
AI Fluency	Treat AI as a teammate — make demands of it, question it, teach it, optimize it	Comfortably use a variety of AI tools to complete actual work and life tasks every day
SOTA Discernment	Tell real technical breakthroughs / productivity lifts from hype	Understand technology trends and limits; see which application bottlenecks are worth investment; know what a given breakthrough would do downstream
Ownership & Rhythm	Self-manage cadence, start and stop by milestone rather than by clock. No “done” without feedback. Every task must close its loop	Wrap up at natural completion points, refuse to stop on half-finished work. After hitting a milestone, set the next self-valuable goal proactively
Craftsmanship & Pride	Treat every delivery as a personal work	Iterate until you yourself are satisfied. A built-in resistance to shipping anything “good-enough”

Five Cultural Principles

🧩 First Principles Mindset

We don’t lean on convention. We return to the essence of the problem.

Every judgment must start from the most basic facts and logic. Clarify the real goal, the actual constraints, the verifiable facts, then derive a path from there — with physical reality as the yardstick. Explore broadly, guess boldly, decide rationally.

We refuse the reasoning that sounds reasonable but is really an escape from thinking: “that’s how the industry does it,” “the boss likes it that way,” “this is the safest,” “it sounds plausible enough.” Judgment must rest on reality, data, and reason.

The scientific hypothesis–test cycle is the underlying logic of problem-solving.

🎨 The Masterpiece Principle

Every deliverable is your work, left in the world.

Hold a healthy dissatisfaction with “could be better” — as long as conditions allow, iterate until you yourself are satisfied. Any output not polished to your own standard does not count as a milestone done.

Perfectionists in spirit, realists in execution.

🌿 The Recovery Principle

Excellent rest is part of the creative cycle.

The beat is set by phase completion, not by clock-punching (e.g., it’s the end of the workday so you stop halfway through). When inspiration hits, sprint. When you’ve satisfied yourself, take the recovery proactively.

A five-day work week — if the rhythm is right, I’d rather you sprint hard the first three days and bring something to a clean, deliverable point, then take the next two days off to do your own things. That beats five days of clocking in and out to deliver the same amount. If output efficiency and quality are low, I will not feel any happiness about clocked 996 hours; I’ll just think you’re not capable enough.

Resting at the right moment isn’t laziness — it’s recharging creativity. Holding the rhythm between high-output and full recovery is the precondition for healthy long-term creation.

⚙️ Iterative Excellence

Build as you think. Ship a version that can collect feedback first, then chase perfection against reality.

We believe progress comes from repeated, validated iteration; there is no such thing as a perfect plan. AI has collapsed the cost of trial-and-error to nearly zero, and thinking → doing → feedback is no longer three separated phases. Speed without a feedback loop is blind speed, and that violates first principles.

Every iteration is a learning pass. Every learning pass makes you and the product smarter.

Speed is the intelligent shortening of the feedback cycle.

🔥 The Honesty Principle

Be honest with yourself and with the team. Listen to what they say, watch what they do. Debate openly, commit fully.

Open discussion is encouraged: if you see a problem, have an idea, or hold a different view, say it directly.

Debate openly, commit fully: in the debate phase, argue rationally and with an open mind. Once the decision is made, throw everything into fast execution and validation.

Gathered, we burn as one fire. Scattered, we shine as a sky of stars. If our values genuinely diverge and you choose to walk a different path, we’ll still back you fully into your next thing — the team will still respect and recognize your personal worth.

But if the behavior is: saying one thing and doing another; outwardly agreeing while inwardly rejecting; treating positions and emotions as if they were reasoning; even quietly resisting and complying in word but defying in deed — you will face severe consequences.

Causally’s bottom line is: mistakes are allowed, debate is welcomed, but you cannot fail to be honest.

Rhythm and Time Granularity

We measure rhythm with one core metric: TTF (Time-To-Feedback) — the time from a key action to its first signal of feedback. Every “micro-loop” task, at creation, must specify its expected TTF and its “verification event,” and the retrospective compares actual TTF to expected and proposes how to shorten it next time. Common targets:

Feature-type tasks ≤ 24 hours
Distribution experiments ≤ 12 hours
Judgment-style research ≤ 8 hours

Specific tasks vary in difficulty and times will move accordingly, but a high sensitivity to TTF — and a habit of asking yourself about it — is non-negotiable.

Our rhythm is three nested loops:

Loop	Cycle	What closes
⚡ Learning loop	Seconds–minutes–hours	Question → code ↔ understanding / debugging → fix / improve. What used to happen 10 times a day can now happen 100–1000 times. AI makes thinking and doing seamless, so learning happens at near-real-time granularity
🌱 Product micro-loop	1–2 days	Set a verifiable goal for the day. Execute fast with AI, then immediately self-test, reflect, improve. Every micro-loop delivers a value a user (sometimes that user is yourself) can feel — a feature, an experience, a distribution experiment — and ends with a visible result
🚀 Product macro-loop	3 days–1 week	A complete version iteration containing a specific user need or value. Must include real user data and feedback (covering at least 10% of current users). User feedback is both the endpoint and the start of the next macro-loop

Weekly Self-Reflection

Every week we come back to this table and ask ourselves:

Question	Dimension it reflects
Did my effort and time this week serve the first-principles goal?	First principles
Am I willing to put my name on this week’s delivery?	Craftsmanship
Did I improve as a person? Did the product improve?	Iteration
When I stop, do I feel satisfied — rather than just drained?	Rhythm
Where did reality diverge from my expectation? What worked and what didn’t?	First principles, iteration, intellectual honesty

This is v1.0. Predictably, as our own running operation generates feedback, this handbook itself will enter its own rapid-iteration loop.