Where are the biggest opportunities in the AI stack?

The biggest opportunities are generally at the application layer, as these are needed to generate revenue to support the underlying foundation, cloud, and semiconductor layers.

What is agentic AI and why is it important for startups?

Agentic AI refers to workflows where an AI system can perform iterative tasks like outlining, researching, drafting, and revising. This leads to better work products and opens up more startup opportunities, especially in orchestrating various AI calls.

How can startups move faster?

Startups can move faster by focusing on concrete product ideas that engineers can build, leveraging subject matter expert intuition ('gut'), pursuing one clear hypothesis at a time, and improving their feedback loops.

How has AI changed software engineering and product development?

AI coding assistance dramatically increases engineering speed and reduces cost, especially for prototypes, shifting the bottleneck to user feedback and product management. This also makes software architecture decisions more like 'two-way doors'.

Should everyone learn to code in the age of AI?

Yes, Andrew Ng believes it's crucial for everyone, regardless of their role, to learn to code. Better coding tools empower more people, and understanding how to instruct computers (even via AI) is a key future skill.

What is the biggest bottleneck in startup development now?

With engineers becoming much faster due to AI, the bottleneck is shifting towards product management, user feedback, and designing what features to build. This is changing traditional PM-to-engineer ratios.

Is AI safety or responsible AI the better concept?

Andrew Ng prefers 'responsible AI' over 'AI safety,' arguing that safety is not a function of the technology itself but how it is applied. Responsible application determines whether AI is beneficial or harmful.

How should founders balance product building with societal downsides?

Founders should look inward and fundamentally assess if their product will make people better off. If not, they should consider not building it, even if financially viable. AI Fund has killed projects on ethical grounds.

Andrew Ng: Building Faster with AI

Y Combinator

Science & Technology3 min read44 min video

Jul 10, 2025|425,558 views|8,253|180

YC Y Combinator

Save to Pod

Key Moments

TL;DR

Execution speed is key for AI startups. Focus on concrete ideas, agentic workflows, and rapid feedback loops.

Key Insights

Execution speed is a primary predictor of startup success, amplified by new AI technologies.

Opportunities are greatest at the application layer of the AI stack.

Agentic AI workflows, which involve iterative thinking and revision, significantly improve work quality.

Concrete product ideas, specified in detail for engineers, enable faster building and validation.

Subject matter experts with strong intuition can make faster decisions than relying solely on data.

AI coding assistants drastically increase engineering speed, shifting bottlenecks to user feedback.

Learning to code and understanding AI empowers individuals across all job roles.

Rapid feedback tactics, from gut instinct to A/B testing, are crucial for product development.

Staying updated on AI advancements and building blocks enables novel application development.

Responsible AI development prioritizes how tools are applied, rather than solely focusing on technical safety.

THE CRITICAL ROLE OF EXECUTION SPEED

Andrew Ng emphasizes that for AI startups, execution speed is paramount and a strong predictor of success. The rapid evolution of AI technology, particularly agentic AI and coding assistants, enables startups to move faster than ever before. AI Fund, where Ng is involved, operates by building startups at a rapid pace, averaging one per month. This hands-on experience highlights the importance of not just observing but actively participating in the building process to understand and leverage these advancements for faster execution.

NAVIGATING THE AI STACK AND THE RISE OF AGENTIC AI

Ng identifies the application layer as the primary area for startup opportunities within the AI stack, above semiconductors, clouds, and foundation models. He highlights agentic AI as a key trend, enabling AI systems to perform tasks more effectively through iterative processes. Unlike linear output generation, agentic workflows involve outlining, researching, drafting, critiquing, and revising, leading to a significantly better final product and opening new avenues for startups focused on orchestrating these complex workflows.

THE POWER OF CONCRETE IDEAS AND EXPERT INTUITION

To achieve speed, Ng advocates for focusing on concrete product ideas that are detailed enough for engineers to build. Vague ideas, while popular, hinder progress. Concreteness allows for rapid development, testing, and validation. He also stresses the value of subject matter experts with strong intuition, who can make faster, more effective decisions than relying solely on data, especially in the early stages of identifying and developing good ideas. This 'gut feeling,' honed by long-term experience, becomes a crucial decision-making proxy.

ACCELERATED ENGINEERING WITH AI CODING ASSISTANTS

AI coding assistants are revolutionizing software development, dramatically increasing engineering speed and reducing costs. This allows for rapid iteration and prototyping, transforming the build-feedback loop. Ng differentiates between building quick prototypes, where AI offers a tenfold speed increase, and production software, where gains are around 30-50%. This shift makes it feasible for startups to build numerous prototypes to test ideas cheaply, reducing the risk of building something nobody wants and embracing a 'move fast and be responsible' mantra.

RESHAPING PRODUCT MANAGEMENT AND DEVELOPMENT WORKFLOWS

As engineering becomes faster, product management and user feedback loops emerge as the new bottlenecks. The traditional PM-to-engineer ratio is shifting, with some teams proposing scenarios where product managers significantly outnumber engineers. This emphasizes the need for efficient methods to gather rapid user feedback. Ng outlines a portfolio of tactics, from personal gut checks to A/B testing, stressing that even slower methods like A/B testing can be accelerated by using data to refine intuition and mental models for quicker, high-quality product decisions.

EMBRACING CODE, AI LITERACY, AND RESPONSIBLE INNOVATION

Ng predicts that learning to code and understanding AI will become essential across all job roles, not just engineering, empowering individuals and boosting productivity. He also stresses the importance of responsible AI development, emphasizing that safety is a function of application, not just technology. While AI offers immense opportunities, founders must consider potential societal downsides and kill projects ethically if they foresee negative consequences. Protecting open-source AI development from excessive regulation is also highlighted as crucial for innovation and equitable diffusion of technology.

Mentioned in This Episode

●Software & Apps

●Companies

●Organizations

●Books

●People Referenced

Common Questions

Andrew Ng identifies execution speed as a strong predictor of a startup's odds of success. Moving quickly allows for rapid validation or falsification of ideas.

Topics

Startup Speed Execution Speed AI Stack Application Layer Concrete Ideas Product Feedback AI Coding Assistants

Mentioned in this video

legislationSB 1047

A proposed bill in California that Andrew Ng fought against, which would have imposed burdensome regulatory requirements on AI development, making it difficult for smaller entities and open-source software.

softwareClaude 3

Currently being used for coding assistance, indicative of the rapid advancements in agentic coding assistants.

softwareJasper

Mentioned as an example of a company that ran into trouble, countering the hype narrative that AI can easily wipe out thousands of startups.

organizationAI Fund

A venture studio that builds startups, focusing on execution speed and concrete ideas in AI.

publicationThe Wall Street Journal