Jay Wang

Data and AI leader, ex-twitter/Microsoft/Snap
RecSys, GenAI, coaching
Creator of , an AI-powered visual search & generation engine.

[Jay Wang @qqwjq]

AI & Data Workflow (6/6)
The real debate isn't Assistant vs. Agent.
It's whether AI will eventually own problem framing, not just execution.
Modern LLMs don't just answer questions.
They propose hypotheses, identify anomalies, ask follow-ups, and run the next analysis.

[Jay Wang @qqwjq]

AI & Data Workflow (5/6)
Data Applications → RT Intelligence
Shift from periodic analytics to cont' decision-making.
Across:
- Search, Ads, Reco
- Autonomous driving
Modular pipelines → End-to-End AI systems
Models are becoming decision-makers, not just prediction engines.

[Jay Wang @qqwjq]

AI & Data Workflow (4/6)

Data Analysis → From SQL to AI Collaboration

LLMs can already:

Write SQL
Generate hypotheses
Build baselines
Create visualizations
Explain results

The analyst evolves from:

Query Writer → Problem Framer + Validator

[Jay Wang @qqwjq]

AI & Data Workflow (3/6)

Data Storage → Vector Infrastructure

Data is no longer accessed only through tables.

Text, images, video, user behavior, and knowledge are all mapped into a shared embedding space.

Consensus:
Vector search + RAG are becoming standard infrastructure.

[Jay Wang @qqwjq]

AI & Data Workflow (2/6)
Data Collection → Synthetic Data
Data is no longer just collected. It's generated.
- Persona & user behavior simulations
- Recommender cold-start data
- RL environments
- AI-generated A/B tests
Consensus: Synthetic data becomes first-class infrastructure

[Jay Wang @qqwjq]

AI & Data Workflow (1/6)
AI isn't just changing data analysis—it's rewriting the entire data workflow.
For the past decade, the pipeline barely changed:
→ Collection
→ Storage
→ Analysis
→ Application
Now every layer is being rebuilt by LLMs and agents.

[Jay Wang @qqwjq]

AI Geopolitics (5/5)
The Space Race wasn't really about the Moon.
It created semiconductors, GPS, satellites, and the modern technology stack.
Likewise, AI isn't ultimately about chatbots.

It's about who builds the strongest technological foundation for the next 50 years.

[Jay Wang @qqwjq]

From one product idea to a complete concept package.

This demo shows an AI-assisted industrial design workflow: simple product requirement to concept sketches, engineering concept boards, exploded views, and presentation-ready marketing assets.
Visit https://t.co/guU9GxZLGC !

[Jay Wang @qqwjq]

AI Geopolitics (1/5)

Open source has become a geopolitical strategy.

China pushes open models like DeepSeek and Qwen to accelerate global adoption.

The U.S. leans on export controls and frontier compute advantages.

[Jay Wang @qqwjq]

AI Geopolitics (3/5)
The competitors are no longer just companies.
The real race is increasing Washington vs. Beijing tension.
The key resources aren't products—they're:
• Chips
• Compute
• Energy
• Talent
• Regulation

[Jay Wang @qqwjq]

AI Geopolitics (2/5)
Sputnik wasn't terrifying because it was a satellite.

It proved the Soviet Union could mobilize science, industry, education, and manufacturing at national scale.

DeepSeek's emergence in 2025 sparked a similar realization.

Many call it AI's Sputnik Moment.

[Jay Wang @qqwjq]

AI Geopolitics (1/5)

The U.S. and China are competing across the entire AI stack, from foundation models to chips & semiconductors, from manufacturing to energy and talents.

AI Isn't the Next Internet.
It's the Next Space Race. 🧵

[Jay Wang @qqwjq]

From one idea to a complete product concept package.
Concept sketches → Engineering boards → Exploded views → Marketing assets.
Looking for design teams and manufacturers interested in testing this workflow. DM us if you'd like to try it with your own product.

[Jay Wang @qqwjq]

AI-Native Collaboration (7/7)

My takeaway:

The next generation of organizations won't be defined by AI agents.

They'll be defined by AI-native collaboration models.

The winners won't simply build faster.

They'll organize better.

[Jay Wang @qqwjq]

AI-Native Collaboration (6/7)

Ironically, management becomes harder.

If everyone can independently build a product.

Who decides priorities?

Who prevents overlapping?

Who integrates everything into one coherent?

AI reduces execution costs.

It increases coordination costs.

[Jay Wang @qqwjq]

AI-Native Collaboration (5/7)

This is why AI-native startups look different.

Fewer specialists.

More end-to-end builders.

More ownership.

Faster iteration.

The bottleneck shifts from execution to judgment.

[Jay Wang @qqwjq]

AI-Native Collaboration (4/7)

The biggest productivity gain isn't coding.

It's eliminating handoffs.

Every meeting, context switch, PRD, and design review carries latency and information loss.

The cheapest communication is the one that never happens.

[Jay Wang @qqwjq]

AI-Native Collaboration (3/7)
A strong builder can now:
- define the problem
- prototype the product
- write production code
- analyze usage
- generate design
- talk to customers
Not coz they're suddenly experts at everything.
Because AI amplifies strengths and fills skill gaps.

[Jay Wang @qqwjq]

AI-Native Collaboration (2/7)
For decades, scaling meant specialization.
PM → Design → Engineering → Data → Marketing.
Every handoff added coordination cost.
Large organizations accepted it because specialization improved execution.
AI changes that equation.

[Jay Wang @qqwjq]

AI-Native Collaboration (1/7)
Everyone knows AI-native startups and large organizations operate differently.
The interesting question isn't why.
It's how AI is changing the economics of collaboration.