Mar 13, 2025 · 34m · a16z

Vibe Coding is Here - How AI is Changing How We Build Online

Steph Smith · 5m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this episode of the a16z podcast, host Steph Smith speaks with partners Yoko Li and Justine Moore about the rapid rise of 'vibe coding'—a paradigm shift where AI agents generate full-stack software from natural language prompts. The panel explores the tools driving this movement, underlying technical architectures, emerging startup business models, and the long-term impact on non-technical creators and software engineering.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The host holds 18.3% of the talking time here. How this is scored →

The host as informed peer 4.3 Guest teaching 5.3 Guest disagreement 1.3 The host pushing back 2.3
05100:0010:0020:0030:000:27–3:56 · The host as informed peer 5/10 What is Vibe Coding? Steph frames the discussion by citing Andre Karpathy coining 'vibe coding' and referencing Reddit communities, setting an informed context. Justine and Yoko outline the basic distinction between technical developers using IDEs and non-technical users using browser tools.3:56–7:52 · The host as informed peer 5/10 Commercial Scale and Latent Demand Steph pushes back on whether these tools build real software or merely static websites. Justine educates Steph by walking through how she built a dynamic web app with OAuth, Supabase, and OpenAI API calls without writing raw code.7:52–10:27 · The host as informed peer 4/10 Technical Foundations: Why LLMs Excel at Coding Steph asks why LLMs perform so remarkably well on code compared to other domains. Yoko delivers a deep technical breakdown covering training data distribution, Stack Overflow, web abstractions, and contained JavaScript runtimes.10:27–15:37 · The host as informed peer 4/10 System Architecture and Execution Environments Steph presses on where these products break down in practice. Justine and Yoko candidly explain context loss in large codebases, regression bugs during minor design tweaks, and the overly optimistic nature of LLM agent debugging.15:37–18:29 · The host as informed peer 3/10 Use Cases from Personal Apps to Agency Workflows Justine and Yoko explore the wide spectrum of use cases, highlighting 'software for one' personal utility apps alongside marketing agencies taking on small client builds that were previously unviable.18:29–22:02 · The host as informed peer 5/10 Product Evolution, Interface Segmentation, and Design Yoko directly rejects the popular premise that everyone should learn to code, arguing code is no longer the correct abstraction layer. Steph counters with a nuanced reframe on degrees of language abstraction.22:02–26:12 · The host as informed peer 6/10 Monetization Strategies and Pricing Dynamics Steph asserts that vibe coding products are severely underpriced relative to the immense value they unlock. Justine and Yoko explain why token consumption pricing is confusing for consumers and propose value-based seat or maintenance models.26:12–29:18 · The host as informed peer 5/10 Startup Defensibility vs Incumbent Platforms Steph challenges the guests on whether startups can survive once incumbents like Google or Figma incorporate AI coding features. Yoko defends startup defensibility by pointing to enterprise inertia and 'shipping your org chart'.29:18–34:30 · The host as informed peer 6/10 The Future of Vibe Coding and Model Context Protocol Steph shares insights on bi-directional prompting dynamics from personal experience and asks Yoko to define the Model Context Protocol (MCP). Yoko breaks down how MCP acts as an extensible tool-use standard for agents.34:30–34:36 · The host as informed peer 0/10 Legal Disclaimer and Disclosures Brief wrap-up and sign-off segment with no substantive debate or technical exchange.0:27–3:56 · Guest teaching 4/10 What is Vibe Coding? Steph frames the discussion by citing Andre Karpathy coining 'vibe coding' and referencing Reddit communities, setting an informed context. Justine and Yoko outline the basic distinction between technical developers using IDEs and non-technical users using browser tools.3:56–7:52 · Guest teaching 6/10 Commercial Scale and Latent Demand Steph pushes back on whether these tools build real software or merely static websites. Justine educates Steph by walking through how she built a dynamic web app with OAuth, Supabase, and OpenAI API calls without writing raw code.7:52–10:27 · Guest teaching 7/10 Technical Foundations: Why LLMs Excel at Coding Steph asks why LLMs perform so remarkably well on code compared to other domains. Yoko delivers a deep technical breakdown covering training data distribution, Stack Overflow, web abstractions, and contained JavaScript runtimes.10:27–15:37 · Guest teaching 6/10 System Architecture and Execution Environments Steph presses on where these products break down in practice. Justine and Yoko candidly explain context loss in large codebases, regression bugs during minor design tweaks, and the overly optimistic nature of LLM agent debugging.15:37–18:29 · Guest teaching 5/10 Use Cases from Personal Apps to Agency Workflows Justine and Yoko explore the wide spectrum of use cases, highlighting 'software for one' personal utility apps alongside marketing agencies taking on small client builds that were previously unviable.18:29–22:02 · Guest teaching 7/10 Product Evolution, Interface Segmentation, and Design Yoko directly rejects the popular premise that everyone should learn to code, arguing code is no longer the correct abstraction layer. Steph counters with a nuanced reframe on degrees of language abstraction.22:02–26:12 · Guest teaching 6/10 Monetization Strategies and Pricing Dynamics Steph asserts that vibe coding products are severely underpriced relative to the immense value they unlock. Justine and Yoko explain why token consumption pricing is confusing for consumers and propose value-based seat or maintenance models.26:12–29:18 · Guest teaching 6/10 Startup Defensibility vs Incumbent Platforms Steph challenges the guests on whether startups can survive once incumbents like Google or Figma incorporate AI coding features. Yoko defends startup defensibility by pointing to enterprise inertia and 'shipping your org chart'.29:18–34:30 · Guest teaching 6/10 The Future of Vibe Coding and Model Context Protocol Steph shares insights on bi-directional prompting dynamics from personal experience and asks Yoko to define the Model Context Protocol (MCP). Yoko breaks down how MCP acts as an extensible tool-use standard for agents.34:30–34:36 · Guest teaching 0/10 Legal Disclaimer and Disclosures Brief wrap-up and sign-off segment with no substantive debate or technical exchange.0:27–3:56 · Guest disagreement 1/10 What is Vibe Coding? Steph frames the discussion by citing Andre Karpathy coining 'vibe coding' and referencing Reddit communities, setting an informed context. Justine and Yoko outline the basic distinction between technical developers using IDEs and non-technical users using browser tools.3:56–7:52 · Guest disagreement 2/10 Commercial Scale and Latent Demand Steph pushes back on whether these tools build real software or merely static websites. Justine educates Steph by walking through how she built a dynamic web app with OAuth, Supabase, and OpenAI API calls without writing raw code.7:52–10:27 · Guest disagreement 0/10 Technical Foundations: Why LLMs Excel at Coding Steph asks why LLMs perform so remarkably well on code compared to other domains. Yoko delivers a deep technical breakdown covering training data distribution, Stack Overflow, web abstractions, and contained JavaScript runtimes.10:27–15:37 · Guest disagreement 1/10 System Architecture and Execution Environments Steph presses on where these products break down in practice. Justine and Yoko candidly explain context loss in large codebases, regression bugs during minor design tweaks, and the overly optimistic nature of LLM agent debugging.15:37–18:29 · Guest disagreement 0/10 Use Cases from Personal Apps to Agency Workflows Justine and Yoko explore the wide spectrum of use cases, highlighting 'software for one' personal utility apps alongside marketing agencies taking on small client builds that were previously unviable.18:29–22:02 · Guest disagreement 5/10 Product Evolution, Interface Segmentation, and Design Yoko directly rejects the popular premise that everyone should learn to code, arguing code is no longer the correct abstraction layer. Steph counters with a nuanced reframe on degrees of language abstraction.22:02–26:12 · Guest disagreement 1/10 Monetization Strategies and Pricing Dynamics Steph asserts that vibe coding products are severely underpriced relative to the immense value they unlock. Justine and Yoko explain why token consumption pricing is confusing for consumers and propose value-based seat or maintenance models.26:12–29:18 · Guest disagreement 3/10 Startup Defensibility vs Incumbent Platforms Steph challenges the guests on whether startups can survive once incumbents like Google or Figma incorporate AI coding features. Yoko defends startup defensibility by pointing to enterprise inertia and 'shipping your org chart'.29:18–34:30 · Guest disagreement 0/10 The Future of Vibe Coding and Model Context Protocol Steph shares insights on bi-directional prompting dynamics from personal experience and asks Yoko to define the Model Context Protocol (MCP). Yoko breaks down how MCP acts as an extensible tool-use standard for agents.34:30–34:36 · Guest disagreement 0/10 Legal Disclaimer and Disclosures Brief wrap-up and sign-off segment with no substantive debate or technical exchange.0:27–3:56 · The host pushing back 1/10 What is Vibe Coding? Steph frames the discussion by citing Andre Karpathy coining 'vibe coding' and referencing Reddit communities, setting an informed context. Justine and Yoko outline the basic distinction between technical developers using IDEs and non-technical users using browser tools.3:56–7:52 · The host pushing back 4/10 Commercial Scale and Latent Demand Steph pushes back on whether these tools build real software or merely static websites. Justine educates Steph by walking through how she built a dynamic web app with OAuth, Supabase, and OpenAI API calls without writing raw code.7:52–10:27 · The host pushing back 1/10 Technical Foundations: Why LLMs Excel at Coding Steph asks why LLMs perform so remarkably well on code compared to other domains. Yoko delivers a deep technical breakdown covering training data distribution, Stack Overflow, web abstractions, and contained JavaScript runtimes.10:27–15:37 · The host pushing back 4/10 System Architecture and Execution Environments Steph presses on where these products break down in practice. Justine and Yoko candidly explain context loss in large codebases, regression bugs during minor design tweaks, and the overly optimistic nature of LLM agent debugging.15:37–18:29 · The host pushing back 1/10 Use Cases from Personal Apps to Agency Workflows Justine and Yoko explore the wide spectrum of use cases, highlighting 'software for one' personal utility apps alongside marketing agencies taking on small client builds that were previously unviable.18:29–22:02 · The host pushing back 4/10 Product Evolution, Interface Segmentation, and Design Yoko directly rejects the popular premise that everyone should learn to code, arguing code is no longer the correct abstraction layer. Steph counters with a nuanced reframe on degrees of language abstraction.22:02–26:12 · The host pushing back 3/10 Monetization Strategies and Pricing Dynamics Steph asserts that vibe coding products are severely underpriced relative to the immense value they unlock. Justine and Yoko explain why token consumption pricing is confusing for consumers and propose value-based seat or maintenance models.26:12–29:18 · The host pushing back 4/10 Startup Defensibility vs Incumbent Platforms Steph challenges the guests on whether startups can survive once incumbents like Google or Figma incorporate AI coding features. Yoko defends startup defensibility by pointing to enterprise inertia and 'shipping your org chart'.29:18–34:30 · The host pushing back 1/10 The Future of Vibe Coding and Model Context Protocol Steph shares insights on bi-directional prompting dynamics from personal experience and asks Yoko to define the Model Context Protocol (MCP). Yoko breaks down how MCP acts as an extensible tool-use standard for agents.34:30–34:36 · The host pushing back 0/10 Legal Disclaimer and Disclosures Brief wrap-up and sign-off segment with no substantive debate or technical exchange.

speaking balance: gold is the host, purple is the guest (3 minute bins)

0:00 · the host 22.7% · guest 77.3%0:00 · the host 22.7% · guest 77.3%3:00 · the host 25.1% · guest 74.9%3:00 · the host 25.1% · guest 74.9%6:00 · the host 16.4% · guest 83.6%6:00 · the host 16.4% · guest 83.6%9:00 · the host 13.6% · guest 86.4%9:00 · the host 13.6% · guest 86.4%12:00 · the host 8.1% · guest 91.9%12:00 · the host 8.1% · guest 91.9%15:00 · the host 8.9% · guest 91.1%15:00 · the host 8.9% · guest 91.1%18:00 · the host 12.7% · guest 87.3%18:00 · the host 12.7% · guest 87.3%21:00 · the host 33.2% · guest 66.8%21:00 · the host 33.2% · guest 66.8%24:00 · the host 19.4% · guest 80.6%24:00 · the host 19.4% · guest 80.6%27:00 · the host 22.6% · guest 77.4%27:00 · the host 22.6% · guest 77.4%30:00 · the host 10.4% · guest 89.6%30:00 · the host 10.4% · guest 89.6%33:00 · the host 32.9% · guest 67.1%33:00 · the host 32.9% · guest 67.1%
Sharpest disagreement ▶ 20:34 Yoko rejects 'everyone should learn to code'

Yoko explicitly dissents from the common tech narrative, stating that code is no longer the right abstraction layer for the majority of people.

Hardest push from the host ▶ 5:31 Steph questions capability limit of vibe tools

Steph directly challenges the assumption that these tools can build real applications, asking if they are limited to basic static web pages.

Biggest teaching moment ▶ 6:02 Justine demonstrates dynamic web app capabilities

Justine educates the host by detailing step-by-step how a non-technical person can incorporate dynamic auth, database storage, and external API calls using browser tools.

The host holds their own ▶ 1:31 Steph grounds the trend in industry taxonomy

Steph demonstrates expertise by attributing the term to Andre Karpathy, identifying active community hubs like Reddit, and categorizing market entrants.

the scores for every segment, with the reasoning behind each
ChapterTopicThe host as informed peerGuest teachingGuest disagreementThe host pushing backWhy
What is Vibe Coding? 5411 Steph frames the discussion by citing Andre Karpathy coining 'vibe coding' and referencing Reddit communities, setting an informed context. Justine and Yoko outline the basic distinction between technical developers using IDEs and non-technical users using browser tools.
Commercial Scale and Latent Demand 5624 Steph pushes back on whether these tools build real software or merely static websites. Justine educates Steph by walking through how she built a dynamic web app with OAuth, Supabase, and OpenAI API calls without writing raw code.
Technical Foundations: Why LLMs Excel at Coding 4701 Steph asks why LLMs perform so remarkably well on code compared to other domains. Yoko delivers a deep technical breakdown covering training data distribution, Stack Overflow, web abstractions, and contained JavaScript runtimes.
System Architecture and Execution Environments 4614 Steph presses on where these products break down in practice. Justine and Yoko candidly explain context loss in large codebases, regression bugs during minor design tweaks, and the overly optimistic nature of LLM agent debugging.
Use Cases from Personal Apps to Agency Workflows 3501 Justine and Yoko explore the wide spectrum of use cases, highlighting 'software for one' personal utility apps alongside marketing agencies taking on small client builds that were previously unviable.
Product Evolution, Interface Segmentation, and Design 5754 Yoko directly rejects the popular premise that everyone should learn to code, arguing code is no longer the correct abstraction layer. Steph counters with a nuanced reframe on degrees of language abstraction.
Monetization Strategies and Pricing Dynamics 6613 Steph asserts that vibe coding products are severely underpriced relative to the immense value they unlock. Justine and Yoko explain why token consumption pricing is confusing for consumers and propose value-based seat or maintenance models.
Startup Defensibility vs Incumbent Platforms 5634 Steph challenges the guests on whether startups can survive once incumbents like Google or Figma incorporate AI coding features. Yoko defends startup defensibility by pointing to enterprise inertia and 'shipping your org chart'.
The Future of Vibe Coding and Model Context Protocol 6601 Steph shares insights on bi-directional prompting dynamics from personal experience and asks Yoko to define the Model Context Protocol (MCP). Yoko breaks down how MCP acts as an extensible tool-use standard for agents.
Legal Disclaimer and Disclosures 0000 Brief wrap-up and sign-off segment with no substantive debate or technical exchange.

Statements from this episode (4)

Assertion Not checkable as stated
Steph Smith: Andre Karpathy coined the term 'vibe coding'
“The term actually comes from Andre Karpathy, right, who is, like, extremely technical, like, is behind the LLMs that are actually helping people vibe code.”
Steph Smith Mar 13, 2025 ▶ 1:38
Assertion Not checkable as stated
Justine Moore: AI coding tools lose context and break complex applications
“As these apps get bigger and bigger and more complex they lose some of the context of, like, what they've already generated and the environment in which they're operating, and, like, you might ask for one design change. This has happened to me many times with …”
Justine Moore Mar 13, 2025 ▶ 12:54
Prediction Not checkable as stated
Yoko Li: The next step for AI coding tools is automated verification
“So I actually think the next step for a lot of these tools is adding, you know, how to verify it still works. Is it adding tests? Is it adding something else? There's a lot to be, you know, desired today, but I have high hopes that it's going to get there pret…”
Yoko Li Mar 13, 2025 ▶ 15:23
Opinion
Steph Smith: AI vibe coding tools are currently significantly underpriced
“I think these different products are so underpriced given the products that people are able to create in terms of, you know, they literally could not create something in the past, and now they can create something of immense value in their life.”
Steph Smith Mar 13, 2025 ▶ 22:29
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 1,000 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.