The AI Daily Brief: RIP Vibe Coding, Feb 2025–Oct 2025

Host: Nathaniel Whittemore (NLW)
Guest: Sean “Swix” Wang
Date: November 3, 2025

Overview

In this episode, NLW sits down with Sean Wang (aka Swix), prominent voice in AI engineering, to dissect the rapid rise—and now, as they argue, the fall—of “vibe coding” as the dominant trend in AI-powered software development. In the lead-up to the AI Engineer Code Summit, they discuss why “vibe coding” captured the zeitgeist of 2025, why both software engineers and organizations are pushing past its limits, and what new paradigms and challenges are emerging in AI-driven coding and enterprise adoption. The conversation covers shifting tool landscapes, role evolution, industry-wide culture changes, and what’s next for effective engineering teams in the age of AI.

Key Themes and Discussion Points

1. What Was “Vibe Coding”—and Why Did It Explode?

• The term “vibe coding” describes a style where non-technical users and developers rapidly prototype and launch applications using AI tools by “going with the flow” rather than following rigorous engineering practices.

• This movement democratized building software, making it easier for non-engineers to create functional apps by leveraging tools like AI code agents and drag-and-drop builders.

• NLW and Swix emphasize that nobody predicted “vibe coding” would dominate 2025:

  "It was not on the radar as the thing that was going to drive all conversations… No one, at least anyone that I saw, was like, this is the year of coding. This is the year of A.I. coding."
  — NLW (05:01)

• Only a few, like Andrej Karpathy, foresaw its significance:

  "The vibe coding. Only Karpathy said that tweet in February."
  — Swix (05:45)

2. RIP Vibe Coding: Disenchantment and the Push Forward

• Swix boldly declares that “vibe coding” has hit its expiration:

  "I declare the end of vibe coding being cool this month. And I think a lot of what we're meaning to discuss at AI Code Summit is what's after vibe coding. How can we avoid the slop and build software that we don't hate..."
  — Swix (05:55)

• Discomfort among software engineers is mounting:

  • Non-technical users hand off “vibe-coded” prototypes expecting them to be production-ready, disregarding the underlying complexities.
  • Technical debt is accumulating as solutions built “by vibe” don’t integrate with established infrastructure, often leading to a complete rebuild.

  "You have to completely rebuild because it doesn't use any of the same tech. I mean, somewhat exaggerating, but... there's a lot of experimentation in just that front."
  — Swix (07:30)

• Even among engineers themselves, the sloppiness of vibe coding is causing friction regarding maintainability, security, and teamwork (“PRs to other people have to clean up”).

3. New Paradigms: Spec-Driven Development & Translation Roles

• The conversation shifts to emerging alternatives:

  • Spec-driven development: A process-focused approach advocated by Amazon and others to bring rigor back to AI-driven software.
  • New “translation” roles may arise, bridging the gap between non-technical “vibe coders” and professional engineers inside organizations.

  "There's a role or at least a function around sort of translating. You know, if you've got all... these folks who are now able to speak with code... it feels like that's, that's a thing that someone could get really good at..."
  — NLW (13:47)

4. Explosion of AI/Agentic Coding Platforms

• 2025 saw an explosion in coding agents and platforms: Claude Code, Bolt, Lovable, Codex CLI, Cognition, Cursor, etc.

• The space is now fragmented—Swix notes:

  • Platforms are converging on universal handoff and more sophisticated collaboration, but seamless integration among tools is still unsolved.

  "I think now the form factors are you have the IDE or VS code extension, you have the web app, you have Slack... The handoff is not worked out yet."
  — Swix (15:35)

• The classic sync/async spectrum is shifting, as asynchronous “agent” tools become much faster, and a new bifurcation is emerging between background commodity tasks and deep, focused “centaur” human/AI collaboration.

  "The sync mode is for the deepest focused and hardest problems where you need the centaur combination of human and AI."
  — Swix (18:01)

5. Enterprise Adoption and Organizational Change

• Engineering departments were initially the slowest to adopt AI agent tools, but that’s changing rapidly as tools mature and value is proven.

• There’s a new focus on enterprise case studies, with representation from major players (Goldman Sachs, McKinsey, Bloomberg, etc.) at the upcoming Summit.

  "When we started... agent audits around the beginning of the year... engineering departments were surprisingly some of the holdouts... it does feel like there has been a major shift over the course of the year."
  — NLW (23:41)

6. Evolving Terminologies: Context Engineering

• Words like “vibe coding” and now “context engineering” mean different things to different audiences. NLW predicts “context engineering” will soon be both a technical discipline and a broad organizational change mindset:

  "I think that context engineering is going to be a term that has a similar bifurcation... It is also now a, a leadership or sort of a change mindset..."
  — NLW (24:56)

7. The 80/20 “Code AGI” Thesis

• Swix puts forth a memorable heuristic:

  "Code AGI will be achieved in 20% of the time of full AGI and capture 80% of the value of AGI."
  — Swix (27:27)

• Coding is a uniquely verifiable, high-leverage task for AI: improvements here generalize to other domains; the feedback loop is tighter thanks to developer/end user overlap, and the infrastructure now laid for code agent tools will spill over into verticals like finance and office productivity.

8. Agent Labs vs. Model Labs: The New Industry Divide

• Application (“agent lab”) companies—focused on delivering user-facing AI agents—are now thriving, in contrast to the previous dominance of “model labs” which prioritized creating foundational AI models.

• Swix frames the current period as “the product era” of AI, with application and agent layer companies racking up revenue and user traction while model labs pivot to infra:

  "If you really just look at what the heck people are actually having PMF with, it's just agents... The Agent Lab is a thesis that isn't quite fully worked out yet, but it's really just the case for building AI companies in a different way than has been in the past."
  — Swix (32:30)

• OpenAI's pivot (“we’re giving up on products, we’re building infra... you should make more money than us on our models”—35:27) signals a new normal.

• Model labs may still launch agent-lab-style products (Anthropic’s Claude Code as an example), but organizational status and culture typically prioritize research over applied engineering.

Notable Quotes & Memorable Moments

• On Vibe Coding's Downfall:

  “I declare the end of vibe coding being cool this month. And I think a lot of what we're meaning to discuss at AI Code Summit is what's after vibe coding…”
  — Swix (05:55)

• On What’s Next in Development Practices:

  “A lot of people are talking about spec driven development as a way forward... The term that has to sort of replace or complement vibe coding hasn't emerged yet, but I can definitely feel it in the air.”
  — Swix (09:25)

• On the 80/20 of Code AGI:

  “The central realization I had was this: code AGI will be achieved in 20% of the time of full AGI and capture 80% of the value of AGI.”
  — Swix (27:27)

• On Emerging Enterprise Trends:

  “When we started... it was very often the case that the engineering departments were surprisingly some of the holdouts. They were the sort of most intransigent around wanting to adopt new systems... it does feel like there has been a major shift...”
  — NLW (23:41)

• On Industry Realignment:

  “You want to build AGI, go join a model lab. You want to build products that serve users and vertical domains, build an agent lab.”
  — Swix (35:56)

Timeline of Important Segments

• 02:13–05:01: Swix and NLW introduce the surprise dominance of AI/vibe coding in 2025 and recap the emergence of major platforms and agents.
• 05:35–09:25: Swix declares the “end of vibe coding,” details discomfort among engineers, and introduces the push toward more structured development (spec-driven).
• 13:47–15:35: Discussion on the evolving roles required for translating between non-technical and engineering workforces; explosion in coding agent variety.
• 15:35–18:01: Breakdown of sync/async spectrum and the shifts in agentic tool interfaces; agent labs acquiring infrastructure capabilities.
• 22:34–24:36: Organizational change, especially in large enterprises, and evolving attitudes towards AI coding adoption.
• 24:56–26:28: NLW’s prediction around “context engineering” as the next buzzword that could split into separate technical and leadership meanings.
• 27:27–32:30: Swix elaborates his 80/20 code AGI thesis and discusses how agentic systems generalize beyond coding.
• 32:30–38:34: Deep dive on agent labs vs. model labs, OpenAI’s shift, and broader industry trends.
• 38:44–End: Closing remarks, thoughts on the importance and perennial challenge of AI ROI measurement, and a preview of upcoming talks and surveys.

Conclusion: What’s Next?

As “vibe coding” loses its luster, the AI engineering world is rapidly iterating on new standards, tools, and cultures to harness AI’s power without succumbing to technical debt or organizational gridlock. The battlefield is shifting toward agentic applications, structured interfaces for non-technical builders, and organizational strategies to capture ROI and productivity. Both NLW and Swix see 2026 heralding deeper integration, evolving terminologies, and a new set of (hopefully improved) problems for the industry to solve.

For anyone looking to understand the cutting edge of AI engineering—both the hype and the hard truths—this episode is a must-listen.