Growing up and going through grad school, I’ve always admired people who possess a high level of thought clarity. And I don’t know where I am on that scale. For some people the principles seem to be just crystal clear, but for me I seem to need to constantly question and re-evaluate things in my mind.
I came across Yan Wang’s context infrastructure and was super curious about the idea of using AI to help analyze my own thought process to distill my core principles. So I gave it a shot and here’s what I’ve learned from 6.5 years industry experience, and from my recent adoption of AI. The result is far from perfect, but it’s a good start.
Some of Yan’s axioms also apply to the DS job family and therefore I reused them. Using the same A (AI/Agentic), T (Technical), V (Verification and Trust), M (Management) and X (Cross-domain) taxonomy, here is my list.
A (AI/Agentic)
All three come from Yan :) I clearly have huge headroom for improvement.
| ID | Statement | Core Idea |
|---|---|---|
| A01 | Yan’s A02 | Detach from execution mechanics and shift toward managing expectations, resources, and systemic boundaries |
| A02 | Yan’s A03 | AI amplifies the user’s existing engineering taste, architectural vision, and domain expertise. |
| A03 | Yan’s A05 | High quality documentation is the only sustainable path to maintain long term agent productivity |
T (Technical)
| ID | Statement | Core Idea |
|---|---|---|
| T01 | Yan’s T05 | Code is a transient implementation detail and cognition is permanent asset. |
| T02 | Yan’s T09 | Capturing high quality data is a valuable product itself |
| T03 | Methodology as infra | Productizing methodologies into tools is the best way for a DS to maintain a technical lever against growing scale. |
| T04 | Causal grounding | Establish counterfactual and answer what would have happened if we had not done something, before drawing conclusions about impact |
| T05 | Heavy tail sensitivity | True systemic risks and opportunity are hidden in the heavy tail of the distribution |
| T06 | Yan’s T02 | Validate the final results rigorously and give the process flexibility |
V (Verification and Trust)
| ID | Statement | Core Idea |
|---|---|---|
| V01 | Verifiability | A system that fails and no one knows why or where cannot be used for decision making |
| V02 | No skipped tests | Unit-tests, lints, and fixers are entry tickets for code check-in. |
| V03 | Single Source of Truth | Your centralized planning document is the contract for what is being built. No scope creep. |
M (Management)
| ID | Statement | Core Idea |
|---|---|---|
| M01 | Meta-skill and Outcome Bias | Solve for classes of problems through general frameworks (Meta-skills) rather than solving single tasks through narrow, brittle instructions |
| M02 | Global Optimization | Governance decisions are not simple classification problems. They are global optimization exercises performed under finite resource constraints. |
| M03 | Currency First Interpretation | Translating statistics into business currency is the only effective language for leadership and strategic alignment. |
| M04 | Minimalism is Correctness | Minimizing code changes and system complexity is the fundamental way to reduce cascading errors, and maintain high-fidelity governance. |
X (Cross-domain)
| ID | Statement | Core Idea |
|---|---|---|
| X01 | Friction efficiency | The goal of governance is not necessarily eradication, but the decision of precise friction that makes the marginal cost of bad actions exceed its expected gain. |
Conclusion
This is a V0 and I hope to self-iterate into someone with a clearer mind. To be honest, I still don’t know where I am on the scale of thought clarity, but doing this exercise was helpful. All in all we just need to improve based on where we are now :)
Also, if your infrastructure allows, I highly recommend you to use Yan’s context infrastructure to distill your own principles. It’s a fun and insightful process.