Introduction: In the AI Era, Speed Is Competitiveness

In this talk, Andrew Ng emphasizes how AI technology is changing and enabling startups to move faster than ever, and how speed is the key to success. He shares lessons learned from building startups, arguing that execution speed is the strongest predictor of startup success.

"I believe execution speed is the single most powerful predictor of a startup's probability of success."

The advancement of AI -- especially the rise of agentic AI -- is helping startups move far faster. Andrew shares best practices that change every 2-3 months, hoping to help the audience move faster and increase their odds of success.

Opportunities in the AI Ecosystem: Focus on the Application Layer

Andrew describes the AI ecosystem as a stack structure:

• Semiconductors
• Cloud (hyperscalers)
• AI foundation models
• Application layer

While media and social media focus on the lower technical layers, the biggest opportunities actually lie in the application layer.

"If you want to build a startup, almost by definition the biggest opportunities are in the application layer."

Every layer holds opportunities, but applications are what generate real revenue and pay the layers below.

Latest AI Trend: The Rise of Agentic AI

Andrew identifies agentic AI as the most important recent trend in AI. A year and a half ago he was giving talks to explain the concept, but now it's been overused as a marketing buzzword, diluting its meaning.

"I want to technically explain why agentic AI matters and why it opens up more startup opportunities."

Traditional LLM Usage vs. Agentic Workflows

• Previously, people asked LLMs to write an essay in one shot (start to finish at once)
• But neither humans nor AI produce their best work in linear tasks
• Agentic workflows involve iterative, step-by-step work:
  1. Write an essay outline
  2. Conduct web research and gather materials as needed
  3. Write a draft
  4. Review, critique, and revise the draft
  5. Repeat

"Running through multiple loops like this is slower, but produces much better results."

Best Practices for Speed: The Power of Concrete Ideas

Andrew emphasizes that at AI Fund, they work exclusively with concrete ideas.

• A concrete idea is one defined clearly enough that an engineer can start building immediately
• Examples:
  • "Optimize healthcare assets with AI" -> too vague
  • "Software that lets hospitals allow patients to book MRI appointments online" -> concrete

"Concrete ideas bring speed. You can quickly determine whether an idea is good or not."

Advantages of Concrete Ideas

• Clear direction lets the team move fast
• Failures are detected quickly, allowing immediate pivots to other ideas
• Vague ideas get praise from everyone but can't actually be built

"Vague is almost always right, but concrete can be right or wrong. Both are fine. What matters is that you find out fast."

Expert Intuition and Fast Decision-Making

Andrew says that the intuition of experts who have deeply studied a problem is extremely valuable for fast decisions.

• Data-driven decisions matter, but gathering data can take a long time
• Expert intuition is an excellent tool for making fast decisions

"Expert intuition can be a surprisingly good decision-making mechanism."

Focus on One Hypothesis at a Time

• Startups have limited resources, so they should focus on one clear hypothesis at a time
• If data shows the idea is wrong, pivot immediately to another concrete idea

"We relentlessly push one thing, and when the world tells us we're wrong, we immediately pursue something completely different with the same relentlessness."

Build-Feedback Loop Innovation and AI Coding Assistants

AI coding assistants have dramatically accelerated engineering speed.

• Writing code used to be difficult and expensive, but now code isn't as valuable as it used to be
• Completely rewriting the codebase three times a month is now feasible

"Software architecture choices are no longer 'one-way doors you can't exit' -- they're more like 'two-way doors you can walk back through if you change your mind.'"

The Importance of Using the Latest Tools

• AI coding assistants (GitHub Copilot, Cursor, etc.) make a huge difference even half a generation apart
• Teams using the latest tools can move much faster

Should Everyone Learn to Code?

Andrew offers the somewhat controversial opinion that everyone in every role should learn to code.

"My team's CFO, HR lead, recruiter, and even front desk staff all know how to code. It makes everyone better at their jobs."

• AI coding assistants are making coding increasingly accessible, so more people should learn
• In the future, the ability to precisely instruct a computer will be the most important skill

Product Feedback: Using Fast and Diverse Methods

Getting user feedback is increasingly the bottleneck in product development. Andrew introduces various feedback methods from fastest to slowest:

1. Use it yourself and decide by intuition (fastest)
2. Get feedback from 3 colleagues or friends
3. Get feedback from 3-10 strangers
4. Demo and get feedback from random people at cafes or hotel lobbies
5. Deploy a prototype to 100+ people for feedback
6. A/B testing (slowest)

"People working in cafes often don't really want to be working, so they're happy to help when we show them our product. I've made countless product decisions in hotel lobbies and cafes."

Upgrading Intuition Through Feedback

• Don't just look at A/B test results to make decisions -- use that data to continuously upgrade your intuition

"If I thought this product name would perform better, but the results say otherwise, it means my mental model of the user is wrong. I need to keep improving my intuition through data."

The Importance of Understanding AI: A Differentiated Competitive Edge

AI is still an immature technology, so teams that deeply understand AI can move much faster and make better decisions than those that don't.

• For example, chatbot accuracy, workflow prompting vs. fine-tuning, voice recognition latency -- accurate technical judgment can solve problems in days, while poor judgment can waste months

"Teams that understand AI well have a clear edge over those that don't."

Conclusion: Speed and Concreteness Are the Keys to Success

Andrew closes by reemphasizing that speed, concrete ideas, and fast feedback loops are the core of startup success in the AI era.

• Focus on concrete ideas
• Fast decision-making and execution are critical
• Accelerate engineering speed with AI coding assistants, and leverage diverse methods for user feedback
• Deep understanding of AI technology is a differentiated competitive advantage

"As a manager, I'm evaluated on both the speed and quality of my decisions, but speed is really important."

Finally, he advises the audience to always stay current with the latest AI trends while maintaining respect for people.