scripod.com

20VC: Codex vs Claude Code vs Cursor: Who Wins, Who Loses | Will All Coding Be Automated - Do We Need PMs | The Real Bottleneck to AGI | The Three Phases of Agents and What You Need to Know with Alex Embiricos, Head of Codex at OpenAI

Overview

Shownote

Highlights

Transcript

Chapters

Pins

20VC: Codex vs Claude Code vs Cursor: Who Wins, Who Loses | Will All Coding Be Automated - Do We Need PMs | The Real Bottleneck to AGI | The Three Phases of Agents and What You Need to Know with Alex Embiricos, Head of Codex at OpenAI

The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch

Feb 21

20VC: Codex vs Claude Code vs Cursor: Who Wins, Who Loses | Will All Coding Be Automated - Do We Need PMs | The Real Bottleneck to AGI | The Three Phases of Agents and What You Need to Know with Alex Embiricos, Head of Codex at OpenAI

20VC: Codex vs Claude Code vs Cursor: Who Wins, Who Loses | Will All Coding Be Automated - Do We Need PMs | The Real Bottleneck to AGI | The Three Phases of Agents and What You Need to Know with Alex Embiricos, Head of Codex at OpenAI

The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch

The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch

Feb 21

Overview Shownote Highlights Transcript Chapters Pins

Shownote

Alexander Embiricos is the Head of Codex at OpenAI, leading the development of the company's flagship AI coding systems that power automated software generation, debugging and developer workflows. Under his leadership, Codex has become one of the most wide...

Highlights

Alexander Embiricos, Head of Codex at OpenAI, shares insights on the evolution and real-world impact of AI coding systems—not as replacements for engineers, but as force multipliers reshaping development workflows, product roles, and enterprise adoption.

02:41

I'm more motivated by winning than by the fear of losing

05:14

Automation of coding tasks will increase demand for engineers in five years

07:17

A PM can step back, collaborate, and raise quality, but these tasks could be done by an eng lead or a designer

12:47

Human lack of ability or inclination to define tasks is the bottleneck

13:05

Codex app is already being used for non-coding tasks

13:52

Giving AI directly to workers helps them understand its potential and integrate it into workflows

18:00

Sales and marketing are more difficult due to increased market competitiveness

18:49

The vast majority of code is written by AI, and people rarely open IDEs

21:39

Codex automatically reviews nearly all code at OpenAI

25:56

Serving models to competitors helps them learn and supports safe AGI delivery

30:52

Daily active users is preferred over weekly active users because it signals deeper integration into daily workflows

32:05

Conversational interfaces like chat or voice will be fundamental, but power users may prefer functional graphical interfaces tailored to their needs

34:10

The best interfaces for Codex are often good for humans too, like filtering test outputs

35:44

Knowledge-work tasks are more valuable for data collection than raw code repositories

36:54

They usually enlist help from other companies for large-scale data campaigns

39:53

Aim for an un-bottlenecked agent to handle full iterative loops without human review

41:56

The evaluation of a model also has a 'vibes' factor, as people prefer to work with models they like

42:43

Having one central AI agent will make it the center of work, and people will be more likely to adopt automation

50:05

To get a job, one should be optimistic about their abilities, build high-quality projects, demonstrate agency and taste, and share these projects

55:50

Ending unlimited usage of Codex cloud caused significant user blowback

Chapters

Motivation Behind Building AI Tools

00:00

Will Coding Be Automated? Why AI Could Create More Engineers, Not Fewer

05:13

Do We Need PMs? The "Undefined" Product Role and When It Matters

07:17

The Real AGI Bottleneck: Human Prompting, Validation, and "Too Much Effort"

08:06

Three Phases of Agents: Coding → Computer Use → Productized Workflows

13:04

Enterprise Reality Check: Security, Permissions, and Safe Agentic Browsing

13:52

Is Inference the New Sales and Marketing?

17:57

What % of Codex Was Written by AI?

18:49

Do OpenAI Use AI for Code Review?

21:33

Is there any stickiness to AI coding tools?

23:31

What Does "Winning" Mean at OpenAI? Mission, Competition, and Moats

28:22

The Future UI: Chat or Voice

32:04

Agent-to-Agent Workflows: Designing for Approvals, Compliance, and Automation

34:10

Do Coding Models Have a Data Moat?

35:39

How does Codex View Data: Will They Build Their Own Mercor and Turing?

36:50

How Does Codex View Consumer: Will They Compete with Lovable?

37:27

Benchmarks vs "Vibes": How People Actually Judge Models

41:56

Cursor's Edge and the Case for Building Your Own Models

42:43

Is SaaS Dead? What Still Defends Value (Humans + Systems of Record)

47:37

Talent Wars and Career Advice for New Engineers in the AI Era

51:28

Guardrails, the Fully AI-Managed Stack, and a 10-Year Vision for Everyone

1:01:03

Transcript

Harry Stebbings: Welcome to 20 Product with me, Harry Stebbings. Now, 20 Product is the monthly show where we sit down with the best product leaders to reveal their tips, tactics, and strategies to scaling the best products and product teams. Now, the real...