Building Faster with AI

Agentic Workflows

When humans write essays, they don't go word-by-word from start to finish. Instead, they research, outline, draft, revise, research more, and iterate until complete. AI output improves dramatically using this same approach, which Ng calls "agentic workflows" (though marketers have since co-opted the term).

A new agentic orchestration layer is emerging between foundational models and applications. This layer helps users efficiently leverage foundational models for better results.

Two-Way Doors and Code Value

Agentic models make code much less valuable than before. Code can now be rewritten and rebuilt much faster, making software architecture less of a "one-way door"—a binding, irreversible decision. While rewriting codebases isn't easy yet, it's becoming increasingly feasible.

When making decisions, it's helpful to categorize them as one-way doors (binding) versus two-way doors (non-binding), considering what AI can help redo.

The Future of Software Engineering

Ng believes software engineering will become more valuable with AI, not less. He compares the current state to programming's evolution from punch cards to digital text—the easier programming gets, the more useful programming skills become.

Everyone should learn to code now that LLMs can help. Being able to ask AI for precisely what you want is immensely valuable, and programming skills are the best way to achieve that precision.

Product development bottlenecks are shifting toward product management and design because engineers have accelerated so much. Product management efficiency isn't improving as quickly as software engineering, so teams should have fewer developers per product manager than before. Engineers with product instincts tend to perform better.

Faster Product Feedback Tactics

Faster product feedback is essential for overall execution speed in the AI era. Ng suggests learning to identify high-traffic areas like coffee shops or hotel lobbies and respectfully asking people for quick feedback.

Execution Speed and AI

Execution speed strongly correlates with startup success. AI can improve execution speed through concrete ideas, rapid engineering, rapid feedback, and understanding AI technology including deep AI concepts.

Distinguishing Hype from Reality

Ng believes large AI companies have hyped several concepts for marketing and fundraising without proper fact-checking:

AGI is imminent
AI will cause human extinction
AI will replace all jobs
AI will casually destroy thousands of startups
Business moats and uniqueness are critical concerns (when opportunities actually outnumber capable builders)
Edtech will take a decade to figure out with AI