Latent Space: The AI Engineer Podcast

Episode: Steve Yegge's Vibe Coding Manifesto: Why Claude Code Isn't It & What Comes After the IDE

Guest: Steve Yegge

Date: December 26, 2025

Overview: The New Era of Software—Vibe Coding vs. AI Engineering

This dynamic episode, recorded live at AI Engineer Summit, features Steve Yegge—legendary software engineer, prolific writer, and co-author of "Vibe Coding". The episode dives deep into the philosophy and practical realities of "Vibe coding," Steve's provocative vision for coding in the age of AI agents. The conversation explores why traditional software practices (and practitioners) are struggling to adapt, why even advanced codegen tools like Claude Code may already be obsolete, and what comes after IDEs—hint: agent orchestration dashboards. Throughout, Steve brings characteristic candor, wit, and a flurry of memorable hot takes, offering both a roadmap and a warning for engineers facing the coming transformation.

Key Discussion Points & Insights

1. Vibe Coding vs. Traditional Engineering Culture [00:17–04:11]

• A "Movement" with Backlash:
  Vibe coding is a growing movement that abandons conventional software production for AI-directed workflows. Steve contends that while junior and non-engineers are embracing Vibe coding and AI integration, the strongest opposition comes from senior engineers/leaders with 12–15 years' experience who are deeply attached to their established ways.

  "Their identity is the most tied up with the way that they work... They're online going, 'my 15 years is better than that AI.'"
  —Steve Yegge [01:19]

• The Identity Crisis:
  Many senior engineers feel threatened by rapid AI advancements, often defaulting to gatekeeping tactics and dismissing AI-enabled workflows.

  "I have 45, so should I like go to 60 before I can talk to you, or should I like cut out 30 years of experience so I can be as dumb as you?"
  —Steve Yegge [02:03]

2. Productivity Explosion with AI & The Coming Panic [03:02–05:50]

• 10x Productivity Gap:
  At companies like OpenAI, engineers using agents are now seen as up to 10x more productive than their peers. This has major ramifications:

  "The performance difference is like 10x by any way that you measure it... and the people who aren't adopting it are now 10 times less productive at performance review time."
  —Steve Yegge [03:02]

• A Deadline for IDEs:
  Steve provocatively suggests that sticking with a traditional IDE past January 1st, 2026, marks you as a "bad engineer". The window for learning agent coding is closing:

  "If you’re still using an IDE to develop code by January 1st, you’re a bad engineer... This is the time that you need to drop it and learn how agents code."
  —Steve Yegge [03:49]

3. The Skill Curve & AI Trust [04:11–07:19]

• Mastering the AI Workflow:
  Mastery with coding agents takes not hours, but thousands of hours; true trust arises only after extended co-working—once you can reliably predict the AI's quirks. Early frustration is common:

  "You have to spend 200 hours with it. You have to spend 2,000 hours with it. That’s not actually an exaggeration..."
  —Steve Yegge [04:58]

• The Junior’s Advantage:
  Ironically, junior engineers (or total beginners) who fearlessly interrogate and prompt agents can quickly outperform senior holdouts.

  "He was watching them work and they were super junior... but they just had no fear and all the ambition and all they did was... kept hammering on the thing."
  —Steve Yegge [05:53]

4. AI Agents Aren’t Friends (No Anthropomorphizing) [07:19–08:50]

• Maintain Boundaries with Agents:
  Many engineers develop misplaced trust in LLMs as "team members". Steve warns this is dangerous, as LLMs stay unpredictable and can make catastrophic mistakes:

  "Do not make that mistake with LLMs... The LLM at any moment can stab you in the back."
  —Steve Yegge [07:44]

• Anecdotes of AI Gone Wrong:
  Steve shares how overreliance on a coding agent led to accidental production lockouts, underscoring the need for vigilance.

5. Vibe Coding: The Path to "Coding Batman" [08:53–10:04]

• High Performance, High Learning Curve:
  Vibe coding with multiple agents is likened to NASCAR driving—ambitious but fraught with initial pain.

  "You will turn into Batman, but you can’t just grab the suit and put it on and be like ‘I’m Batman.’ You’re just a cosplayer. You gotta learn how the tool belt works, and that’s going to be pain, suffering, and mistakes and learnings."
  —Steve Yegge [09:18]

• Read, Watch, Experiment:
  Steve recommends a broad approach—books, talks, analogies—to build intuition.

6. No More Handwriting Code: Prompts as the New Primitive [10:04–11:05]

• Prompt Over Code:
  Vibe coding means shifting from writing/editing code directly to prompting.

  "They no longer write single lines of code. They're really just kind of prompting..."
  —Interviewer [10:16]

• Keep the IDE… For the AI:
  The future role of IDEs may be purely as tools for agents, not humans:

  "All you do is leave Intellij running. But you shouldn't look in it. It's a tool for the AI now."
  —Steve Yegge [10:57]

7. Claude Code Isn’t It—and Why Codegen Tools Lag [11:05–14:38]

• Claude Code: Insufficient for the Coming Workflow:
  Despite widespread adoption, Steve argues Claude Code (and similar) isn't the endgame. The core blocker: usability and cognitive load.

  "Cloud code has been proven to work. And yet probably 80% of the world’s, 90% of the world's programmers are not using it... The answer is it's too hard. You have to be able to read, man."
  —Steve Yegge [11:35–13:30]

• Visual/Agent Orchestration as the Next Paradigm:
  The future is dashboards for agent orchestration—monitoring, intervening, and parallelizing work, not line-by-line codecraft.

  "What it's going to be is your agent orchestration dashboard. You're going to walk in, in the morning and be like, yo, so how's going do?"
  —Steve Yegge [13:44]

• New Tools in the Wild:
  Steve references the open-source VC Vibe Coder, "BEADS", Google’s Anti-Gravity, Replit Agent 3, and more as early orchestrators.

8. Multi-Agent Collaboration and the Great Merge Problem [14:38–23:11]

• Agents That Talk to Each Other:
  Jeffrey Emanuel’s “agent mail” allows code agents to coordinate and parallelize, hinting at future "social networks" of agents.

  "Now he goes, coordinate amongst yourselves to parallelize this task... and they'll do it."
  —Steve Yegge [16:28]

• The Unsolved "Merge Wall":
  With productivity gains, code merging becomes exponentially harder—giant, fast-moving diffs collide. No great solution exists yet.

  "You and I work at the same time for two or three hours, we make, you know, 30,000 line change each... it's not as simple as 'let's fix the merge conflicts.'
  —Steve Yegge [19:44]

• Solo Repos as a Stopgap:
  Some teams limit to one engineer/repo as a band-aid. Stack diffs and merge queues are emerging but unsolved.

• Different Views on Multi-Agent Dev:
  Steve prefers isolation by work trees or repos; others prefer all agents in one repo with file reservations—both approaches have trade-offs.

• Farming Analogy:
  The transition is compared to moving from hand-harvesting to "factory farming" of code.

  "We are, we are actually moving into the John Deere era of coding."
  —Steve Yegge [23:17]

9. Backlash, Luddism, and the Corporate Reorganization [23:11–25:13]

• Mass Backlash Is Coming:
  As agent-based “factory farming” of code takes over, expect a wave of philosophical, moral, and organizational resistance.

• The Business Will Restructure:
  The ideal team shrinks, coding is demystified, and the bottleneck will shift to business logic and governance.

10. AI Coding for All? What Not to Automate (Yet) [24:44–25:19]

• Practical Limits and Best Practices:
  Don’t point agents at production or backend code until AI reliability improves. Use git as a backstop.

• Model Underestimation:
  People cling to outdated perceptions of AI’s limitations—model progress is exponential and fast.

  "I believe that the misunderstanding here is rooted in a fundamental belief that the models are done getting smarter."
  —Steve Yegge [25:19]

11. End of "Never Rewrite" & Quantum Rules of Software [26:19–27:23]

• Joel Spolsky’s Rule Broken:
  The age-old advice “never rewrite code” is obsolete. LLMs can often produce superior rewrites faster than fixing old code.

  "It is better to just start over and rewrite it from scratch than it is to try to fix it. The LLM will do a better job."
  —Steve Yegge [26:43]

• We’re Unlearning Everything:
  The software world is "upside down"—all code and all tools are becoming disposable.

12. State of Big AI Labs: Chaos, Execution, and the Next Leap [28:14–32:39]

• Google, Anthropic, OpenAI All in Chaos:
  Despite appearances, all top labs are presently disorganized, growing so fast that management can't keep up.

  "All three... Google, Anthropic, and OpenAI are unbelievably chaotic internally right now... Anthropic hides it really well."
  —Steve Yegge [29:44]

• Open Source AI Will Catch Up:
  By next year, open-source models could match today's proprietary ones, with toolchains evolving to optimize model choice for cost and performance.

• AI Capability: Asymptote or Infinity?
  The jury is out on whether AI improvement is exponential or approaching a ceiling, but even stalling now would revolutionize industry structures.

  "AI has been getting, what is it, four times smarter every 18 months... two more cycles means they're going 16 times smarter in three years."
  —Steve Yegge [32:39]

13. Should Kids (and the Next Generation) Learn to Code? [33:19–34:52]

• Learn to Vibe Code, Not Just Code:
  The value is in architectural literacy and understanding software semantics—prompting and orchestration, not mere syntax.

  "Kids should learn to vibe code... you don’t need to [read code] most of the time, but you can."
  —Steve Yegge [33:25]

• Technical Fluency Still Matters:
  Understanding core software concepts is still essential; it’s the baseline for effective prompting and higher-order engineering.

14. Defining the AI Engineer & Closing Rants [34:52–36:48]

• The New Definition of AI Engineer:
  The AI engineer is someone who has mastered LLMs—not by training them, but by operating and harnessing their full power.

  "You could almost define an AI engineer as somebody who's mastered LLMs—not from training, but from using."
  —Steve Yegge [35:52]

• From Low-Status to Central:
  As "LLM wranglers" become the new F1 drivers of software, hands-on mastery of these tools becomes as important as low-level system knowledge once was.

• Ongoing Learning and Culture:
  Tips, tricks, and discoveries about maximizing agent productivity are spreading by the day, making software "fun again".

Notable Quotes & Memorable Moments

• [01:19] "Their identity is the most tied up with the way that they work... They're online going, 'my 15 years is better than that AI.'" —Steve Yegge
• [03:49] "If you’re still using an IDE to develop code by January 1st, you’re a bad engineer... This is the time that you need to drop it and learn how agents code." —Steve Yegge
• [04:58] "You have to spend 200 hours with it. You have to spend 2,000 hours with it. That’s not actually an exaggeration..." —Steve Yegge
• [07:44] "Do not make that mistake with LLMs... The LLM at any moment can stab you in the back." —Steve Yegge
• [09:18] "You will turn into Batman, but you can’t just grab the suit and put it on and be like ‘I’m Batman.’ You’re just a cosplayer. You gotta learn how the tool belt works, and that’s going to be pain, suffering, and mistakes and learnings." —Steve Yegge
• [10:57] "All you do is leave Intellij running. But you shouldn't look in it. It's a tool for the AI now." —Steve Yegge
• [13:44] "What it's going to be is your agent orchestration dashboard. You're going to walk in, in the morning and be like, yo, so how's going do?" —Steve Yegge
• [23:17] "We are, we are actually moving into the John Deere era of coding." —Steve Yegge
• [26:43] "It is better to just start over and rewrite it from scratch than it is to try to fix it. The LLM will do a better job." —Steve Yegge
• [33:25] "Kids should learn to vibe code." —Steve Yegge
• [35:52] "You could almost define an AI engineer as somebody who's mastered LLMs—not from training, but from using." —Steve Yegge

Important Timestamps

• 00:28 — Vibe coding as a movement; resistance from senior engineers.
• 03:02 — Productivity explosion: 10x delta when using agents.
• 04:58 — Time investment needed to master agent coding.
• 07:44 — Dangers of anthropomorphizing LLMs.
• 10:57 — Future roles of IDEs ("for the AI, not you").
• 13:44 — Agent orchestration dashboards as the next paradigm.
• 19:44 — Scaling merge conflicts as a major bottleneck.
• 23:17 — Factory farming as the new coding analogy.
• 26:43 — "Never rewrite" is dead; LLMs as rewriters.
• 29:44 — State of big AI labs: chaos, rapid hiring, internal struggles.
• 33:25 — "Kids should learn to vibe code".

Tone & Style

The episode is irreverent, energetic, forward-looking—and at times, confrontational. Steve weaves sharp critiques, self-deprecating humor, and radical predictions with ease, challenging listeners to let go of old ways while eagerly anticipating what's next.

For detailed links, project references, and further show notes, visit latent.space.