Augmenting Data Science in the AI Era
What should we do, when code has become 1,000 times cheaper?
Presenter: Dr. Yuzheng Sun, Dr. Yan Wang
Date: 1.16.2025, @ Pinterest
My Vantage Point: Living and Breathing AI
As a former data scientist, my full-time role now centers on teaching AI courses and cultivating a thriving AI community. This unique position grants me daily immersion in the evolving world of artificial intelligence.
1
Two Years of Immersion
Our community has been actively running for over 1.5 years, constantly adapting to AI's rapid advancements and challenges.
2
2000+ Tech Alumni
We proudly count over 2,000 alumni from major tech companies among our ranks, contributing diverse perspectives.
3
8000+ Community Members
A diverse and active community of over 8,000 individuals provides a rich pool of wisdom and daily AI-related questions.
Drawing from this collective intelligence, I'm here to share some strong opinions from a data scientist who has been living and breathing AI for the past two years.
The Challenge
The Identity Crisis of Data Science
What is the actual "product" of data science?
True Output
Better Decisions — not models, dashboards, or p-values
The Problem
Decisions are ad-hoc, human-dependent, hard to quantify
The Result
No sustainable moat when process relies on manual, one-off efforts
The AI Shift: From Replacement to Augmentation
It's Not AI vs. DS
It's DS + AI vs. The Old Way
1
Stop Asking
"Will AI replace me?"
2
Start Asking
"How does AI solve historical bottlenecks of my craft?"
3
The Vision
From "AI-in-the-loop" to AI-Native workflows

What should we do, when code has become 1,000 times cheaper?
Foundation
The Three Characters of the Future Data Scientist
Roles will collapse, but characters will be more prominent than ever
1
The Architect
Designing the flow from raw information to actionable decision
2
The Auditor
Guardian against hallucinations — both AI and human
3
The Full-Stack Builder
From "Here's a slide" to "Here's a working prototype"
Character 1
The Insight & Data Architect
Organizing Information for Decision Flow
01
Organize “Data”
Quantitative metrics and Qualitative context
02
Build Infrastructure (Context Engineering)
Create paths where insights flow naturally into business actions
03
Taking Bets
Recommend decisions with uncertainties
Character 2
The Auditor of Truth
Combatting Hallucinations at Scale
AI Hallucinates
Models generate plausible but incorrect outputs
Humans Hallucinate Too
Cognitive biases, logical fallacies, confirmation bias
DS as Truth Verifier
Establishing ground truth as foundation for all AI-generated logic
The Data Scientist evolves into the guardian of relevance in an AI-driven world
Character 3
The Full-Stack Builder
Closing the Gap Between Insight and Impact
Analysis Isn't Enough
We must build functional solutions
Create Prototypes
Demos that stakeholders can touch, test, experience
Product Ownership
Collaborate as co-creators, not ticket-takers
Our Unique Taste: The Pursuit of Truth
Why Data Scientists are Irreplaceable in the AI Loop
Technical Skills
Can be learned, taught, automated
"Taste for Truth"
Cultural DNA — sensitivity to data integrity nuances
The Rarest Commodity
In an AI-native world, truth becomes the most valuable asset
The Barrier
The Impossible Triangle
Why Truth Has Been Hard to Find
Traditionally, you could only pick two:

The fundamental constraint that has limited data science impact for decades
Seeing it in Action
How to use AI to get all three.
The Question
How do we actually implement this vision?
The Answer
A code-first approach to building clear, verifiable data products
Handing over to Dr. Wang
Demonstrating how AI solves the Impossible Triangle and drives 10x better observability for Data Scientists