Andrew Ng Deep Dive: New Business Logic in the AI Era—Why Technology Isn’t the Bottleneck

1. AI Progress: Multiple Paths Beyond Scale

Andrew Ng argues that future AI breakthroughs will arise from several parallel avenues—agentic workflows, multimodal model construction, extensive application engineering, and novel technical breakthroughs—rather than merely scaling model size, which is becoming increasingly difficult and is over‑emphasized by a few well‑publicized companies.

He highlights that diffusion models used for image generation might also be applied to text generation, illustrating the excitement around cross‑modal research.

2. Agentic AI – From Coined Term to Market Hype

When asked why he created the term "Agentic AI," Ng explains that his team initially resisted, but he persisted because debates about what counts as an AI agent were consuming valuable time. He views agent autonomy as a spectrum: at one extreme, highly autonomous agents can plan, perform multi‑step reasoning, and execute complex tasks; at the other, lower‑autonomy systems merely invoke large language models and reflect on outputs.

He suggests abandoning the binary "is it an AI agent?" debate in favor of acknowledging varying degrees of agency and focusing on building functional systems. He notes that marketers later seized the term, inflating hype faster than genuine commercial progress.

3. The Real Bottleneck: Talent, Not Technology

Ng identifies talent as the primary obstacle to scaling Agentic AI workflows. Teams that lack systematic evaluation and error‑analysis processes waste time on random experiments, while those with rigorous engineering practices can move faster.

From a component perspective, AI agents’ ability to control computers is still flaky, and safety guardrails and evaluation remain major challenges. Moreover, many workflow steps require external knowledge that resides in human brains.

He stresses that unless we can build AI avatars that interview employees or visual AI that understands screen content, full automation will remain elusive.

4. Programming AI Agents: The Most Mature Application

According to Ng, AI‑assisted programming tools represent the clearest and largest economic opportunity today. He cites two dominant domains:

Answering user questions—where OpenAI’s ChatGPT leads the market.

Programming AI agents—where he personally prefers Claude Code as a developer tool.

Claude Code can autonomously generate task lists and execute them step‑by‑step, making it one of the most autonomous and effective AI agents in practice.

Success factors he lists include engineers’ ability to integrate components, the obvious economic value of programming, massive resource investment attracting talent, and the fact that developers are the end users, giving them strong product intuition.

5. AI‑Assisted Coding Is a High‑Intensity Intellectual Activity

Ng rejects the notion of "vibe coding" and argues that AI‑assisted coding demands deep, focused thinking. After a full day of AI‑augmented development, he feels mentally exhausted, describing it as "fast engineering"—still engineering, but at unprecedented speed.

6. Startup Bottleneck Shift: From Engineering to Product Management

At AI Fund, Ng observes that rapid engineering and AI‑assisted coding have compressed development cycles: tasks that once required six engineers three months now finish over a weekend.

The core iteration loop bottleneck has moved to product management—deciding what to build—because coding speed and cost have improved while decision‑making remains slow.

Teams now rely heavily on intuition and deep customer empathy to make rapid product decisions.

7. Product‑Management Automation Remains Limited

Various tools aim to accelerate product‑management workflows (e.g., Figma’s design integration, AI‑driven user‑interview assistants, AI agent clusters simulating user groups), but their impact is modest compared with the dramatic speed gains seen in programming tools for engineers.

8. Founder Profile: Technical Depth Trumps Pure Business Experience

Ng asserts that in the fast‑moving AI era, deep knowledge of frontier technologies is the scarcest resource. Founders who are technically proficient and have strong AI intuition are far more likely to succeed than those who rely solely on business acumen without a clear grasp of AI direction.

Without a deep understanding of technological limits, strategic thinking and leadership become difficult.

9. Common Traits of Successful Founders

Beyond technical expertise, Ng identifies four key traits:

Sharp insight into emerging technologies and early recognition of opportunities.

Relentless work ethic and a willingness to “go crazy” believing they can change the world.

Dual motivations: a desire for commercial success and a genuine obsession with helping customers succeed.

Rapid decision‑making ability, likened to reacting instantly in a tennis match, requiring deep knowledge and intuition.

10. The "Small‑But‑Sharp" Team Model Powered by AI

Team‑size trade‑offs have shifted: previously, large teams outsourced work to cut costs; now, AI assistants enable tiny, highly skilled teams equipped with many AI tools to outperform larger, unevenly capable groups.

The most efficient teams are often the smallest.

Mindset shift: hire AI rather than add headcount, and cultivate "AI intuition"—the ability to ask for budget to employ an AI for a specific task.

11. Outlook for the Next Five Years

AI is a "rich mine of opportunity" with many unexplored ideas. Concrete concepts are more valuable than macro analysis, and economists studying which jobs are most at risk of AI disruption can uncover promising project ideas.

VC work can be partially automated—deep company research and competitive analysis are suitable for automation, while human judgment still dominates background checks and founder character assessment.

Ways to help founders include sharing industry intuition, assisting with talent recruitment and community building, and providing guidance on financing, customer feedback, and technology trends.

Non‑consensus judgments: in a few years, many people will be dramatically empowered by AI, amplifying personal capabilities beyond current imagination; both individuals and enterprises will become far stronger than they can presently envision.