Deutschภาษาไทย中文
Sequoia AI Ascent 2026 Deep Dive (2/3)

Software 3.0 — From Vibe Coding to Agentic Engineering

Andrej Karpathy's Software Revolution Manifesto

Independent Research | LittleX Research Lab | 2026-05-02
Series theme: AGI is not the future — it's now, and you have an 18-month window

In December 2025, AI godfather Andrej Karpathy started vibe coding. He stopped correcting AI-written code and couldn't even remember the last time he made a fix. A few months later, he took the stage at Sequoia AI Ascent 2026 and said something that silenced the room:

"You can outsource your thinking, but you can't outsource your understanding."

This isn't motivational fluff. This is the definitive judgment from someone who co-founded OpenAI, built Tesla Autopilot, and shaped an entire generation of deep learning — delivered after personally experiencing the AI programming revolution. This talk has surpassed 31 million views on YouTube, making it the most-watched session of the entire conference.

Abstract

This article provides an in-depth analysis of Andrej Karpathy's complete talk at Sequoia AI Ascent 2026. Karpathy introduces the Software 3.0 paradigm — LLMs are not faster tools, but an entirely new kind of computer, where programming shifts from "writing code" to "writing context." Using his personal experience with Menu Gen — an app rendered obsolete by a single Gemini prompt — he demonstrates that entire application layers may not need to exist. He distinguishes between Vibe Coding (raising the floor, enabling everyone to build software) and Agentic Engineering (raising the ceiling, giving professional engineers 10x+ acceleration), and proposes a "verifiability framework" explaining why AI makes blazing progress in math and code yet tells you to walk to a car wash 50 meters away. The deepest insight: understanding cannot be outsourced — it is humanity's final moat in the AI era.

Software 3.0 Vibe Coding Agentic Engineering Verifiability Jagged Intelligence LLM as Computer Context Window Andrej Karpathy Sequoia AI Ascent Ghosts vs. Animals
31M+
YouTube views (highest at the conference)
3
Software paradigm revolutions (1.0 → 2.0 → 3.0)
> 10x
Productivity multiplier for agentic engineers
10-15
Parallel AI workflows Boris Cherny manages simultaneously
Table of Contents
  1. Software 1.0 → 2.0 → 3.0: Three Software Revolutions
  2. Vibe Coding: When Programming Becomes Essay Writing
  3. Agentic Engineering: Where Is the Ceiling?
  4. The Verifiability Framework: What Can You Hand to AI? What Can't You?
  5. One Engineer Orchestrating 15 AI Workflows
  6. "Understanding Cannot Be Outsourced" — Karpathy's Warning
  7. Historical Parallels: Who Gets Eliminated in Every Tool Revolution?
  8. Business Implications: How Companies Should Redefine "Engineer"
  9. Conclusion: Which Side Are You On?
  10. Series Guide
  11. References

I.Software 1.0 → 2.0 → 3.0: Three Software Revolutions

To understand what Karpathy is saying, you first need to understand three fundamental shifts in software. These aren't version upgrades — they are essential changes in the computing paradigm.

Software 1.0: Humans Write Rules

From the 1950s to today, this remains dominant. Engineers use C, Python, Java, and other languages to write precise instructions line by line. The computer executes them strictly, with zero "understanding." Whatever you want it to do, you must tell it exactly how to do every step.

Software 2.0: Machines Learn by Curating Datasets

Karpathy coined this concept in 2017. The core shift: programming is no longer about writing code, but about curating datasets and designing training objectives. You don't tell the computer how to recognize a cat — you give it a million cat photos and let the neural network learn on its own. Programmers become data curators and neural network architects.

Software 3.0: "Programming" in Natural Language

This is what's happening in 2025-2026. Karpathy put it bluntly in his talk:

"Software 3.0 is — your programming now becomes the prompt, and the context window is your joystick for this interpreter (the LLM)."
— Andrej Karpathy, Sequoia AI Ascent 2026

He used the installation process for OpenClaw (an AI coding tool) as an example: in the traditional mindset, you'd expect a shell script to install it. But shell scripts become extremely complex when dealing with different platforms and computer configurations. Under the Software 3.0 paradigm, the installation method becomes a text description — you copy and paste it to your AI agent, and the agent intelligently executes the installation, debugs errors, and keeps going until everything works, adapting to your environment and machine.

First Principles

The core difference across the three eras isn't speed — it's the leap in abstraction level.

  • 1.0: You dictate every step to the machine (How)
  • 2.0: You define the objective, the machine learns the method (What + Data)
  • 3.0: You describe the intent, the machine understands and executes (Why + Context)

With each leap, the details humans need to control decrease by an order of magnitude, but the judgment and understanding required increase by an order of magnitude.

The Menu Gen Paradox: Entire Application Layers May Not Need to Exist

Karpathy shared an example that shocked even himself. He spent considerable time vibe coding an app called Menu Gen — photograph a restaurant menu, OCR the dish names, use image generators to produce pictures of each dish, deploy on Vercel. A complete Software 1.0 + 2.0 hybrid.

Then he saw the Software 3.0 version: just toss the menu photo to Gemini with a one-line instruction, and the AI renders images of each dish directly onto the original photo. No app needed, no backend, no deployment.

"All of my Menu Gen is redundant. It operates in the old paradigm — this app shouldn't even exist."
— Andrej Karpathy
Core Insight

Software 3.0 doesn't help you write code faster — it makes certain programs unnecessary altogether. This is an existential warning for every SaaS founder: is your application merely glue between two LLM capabilities? If so, your moat is zero.

II.Vibe Coding: When Programming Becomes Essay Writing

Karpathy coined the term "Vibe Coding" in early 2025, and by the time he took the Sequoia stage in 2026, he reflected on how the concept had evolved.

He described his personal experience: December 2025 was a clear inflection point. Before that, AI coding tools would write code that "sometimes had errors you'd need to fix." But by December, paired with the latest models, "the code was just correct, I kept asking for more, it kept being correct, and I couldn't remember the last time I made a fix."

And then he started vibe coding — the audience erupted in laughter.

What Is the Essence of Vibe Coding?

"Vibe coding is about raising the floor — enabling everyone to build things with software. The floor has risen, everyone can vibe code anything, and that's just incredible."
— Andrej Karpathy

The key word: floor. Vibe coding enables people with zero programming knowledge to create software applications. A marketer can build a data dashboard; a designer can create an interactive prototype; a retired teacher can build a personal knowledge base. This is democratization, universalization — a wholesale raising of the floor.

But Karpathy then pivoted to point out vibe coding's dark side:

Key Distinction

Vibe Coding lowers the barrier to "doing," but doesn't lower the barrier to "doing well." Anyone can use AI to write a working website, but building a secure, efficient, maintainable system still requires deep engineering understanding. That's why Agentic Engineering is needed.

III.Agentic Engineering: Where Is the Ceiling?

If Vibe Coding raises the floor, then Agentic Engineering raises the ceiling. Karpathy drew a clear distinction between the two:

Vibe Coding (Floor)
  • Enables everyone to build software
  • Rapid prototyping, personal projects
  • Lower quality requirements
  • "Good enough to work"
Agentic Engineering (Ceiling)
  • AI-accelerated professional software engineering
  • No compromise on quality standards
  • No tolerance for security vulnerabilities
  • "Fast AND good"
"Agentic engineering is an engineering discipline. You have these agents, they are spiky entities — somewhat error-prone, somewhat random, but extremely powerful. The question is: how do you coordinate them to move fast without sacrificing quality standards?"
— Andrej Karpathy

The 10x Engineer Myth Is Outdated

The software industry has long talked about "10x engineers" — top engineers being 10 times more productive than average ones. Karpathy believes the agentic era has amplified this gap even further:

"People used to talk about the 10x engineer. I think that multiplier has now been drastically amplified. 10x is not the speedup you can get. What I'm seeing is that people who are really good at this are beyond 10x — from what I can see."
— Andrej Karpathy

What Does an Agentic Engineer Look Like?

Karpathy described his vision of the agentic engineer:

Hiring Must Change Too

Karpathy was blunt: most companies' hiring processes are stuck in the old paradigm:

"If you're still giving little puzzles, that's the old paradigm. Hiring should become: give a large project, see how the candidate implements it. Like build a Twitter clone, deploy it live, and then I send 10 AIs to try to hack your system — they shouldn't be able to."
— Andrej Karpathy
Business Insight

A new dimension of capability gaps. In the agentic era, the gap between engineers is no longer about "coding speed" but about "the ability to orchestrate AI." Using the same Claude Code, one person might only have AI write a single function, while another orchestrates 15 parallel workflows to complete an entire system. The latter's productivity could be 100x the former's. This poses fundamental challenges to organizational design, compensation structures, and performance evaluation.

IV.The Verifiability Framework: What Can You Hand to AI? What Can't You?

This is the most analytically rigorous part of Karpathy's talk. He proposed a simple yet powerful framework for understanding the uneven distribution of AI capabilities:

Karpathy's Verifiability Framework

Traditional computers easily automate "things you can specify in code." LLMs easily automate "things where you can verify the output."

Automation speed = f(verifiability + lab attention)

Why Is AI So "Jagged"?

Karpathy used an example that had the entire audience laughing:

"Opus 4.7 can simultaneously refactor a 100,000-line codebase, or discover a zero-day vulnerability — but it'll tell you to walk to a car wash 50 meters away. That's insane."
— Andrej Karpathy

Why does this happen? Karpathy's explanation involves two factors:

Factor One: Verifiability Determines RL Effectiveness

Frontier labs, when training LLMs, are essentially doing large-scale reinforcement learning (RL). RL requires "verification rewards" — do it right, get positive feedback; do it wrong, get negative feedback. In code, verification is natural: tests pass means correct, compilation fails means wrong. Same with math: the answer is either right or wrong.

But "should you drive or walk to a car wash 50 meters away"? How do you verify that? There's no test to run, no clear right or wrong — just common sense. And common sense happens to be outside RL's training distribution.

Factor Two: Lab Attention Allocation

Karpathy revealed an important but rarely discussed fact: AI capability distribution depends not just on technology, but on what labs choose to focus on. He cited the massive improvement in chess ability from GPT-3.5 to GPT-4 — not because the model got generally stronger, but because "someone at OpenAI decided to include a ton of chess data in the pretraining set."

"If you're in the RL loop, you fly. If you're outside the data distribution, you struggle. You have to figure out which loop your application is in."
— Andrej Karpathy

The Verifiability Spectrum

Domain Verifiability AI Automation Speed Reason
Code Very High Very Fast Tests pass / compilation succeeds
Math Proofs Very High Very Fast Formal verification
Legal Documents Medium-High Fast Cross-verifiable via precedents and statutes
Writing Medium Medium Can use LLM jury, but highly subjective
Design Aesthetics Low Slow No RL reward function for aesthetics
Common Sense Very Low Very Slow "Walk 50m to a car wash" can't be verified
Insight for Founders

If you can create a verifiable RL environment for your domain, you have a technological lever. Even if frontier labs aren't paying attention to your field, you can use your own fine-tuning and RL environments to bring AI to expert-level in your vertical. Karpathy hinted he knows such opportunities exist but didn't want to reveal specifics on stage — "I don't want to vague post," but the audience laughed, understanding the implication.

V.One Engineer Orchestrating 15 AI Workflows

Karpathy's talk was primarily conceptual, but at the same Sequoia AI Ascent 2026 conference, Claude Code creator Boris Cherny's hands-on case study perfectly validated Karpathy's theory.

Boris Cherny's Working Style

Boris Cherny shared at the conference how he works with Claude Code: one engineer simultaneously orchestrating 10 to 15 parallel AI workflows. This isn't theoretical — it's his actual daily working state.

Role Shift
From "person who writes code" to "commander orchestrating an AI army"
Core Competencies
Task decomposition, context design, quality control, parallel coordination
Productivity Multiplier
One person completing what previously required an entire team
New Bottleneck
Human understanding and attention, not AI compute

Context Is the New Code

Both Karpathy's and Cherny's practices point to the same conclusion: context is the code of the Software 3.0 era.

In the past, engineers invested time mastering a programming language's syntax and frameworks. Now, the highest-return investment is:

Karpathy even said his biggest frustration when using third-party libraries isn't code complexity, but that "documentation is still written for humans":

"Why is anyone still telling me what to do? I don't want to do anything. What I want is — what's the text I can copy-paste to my agent?"
— Andrej Karpathy (audience laughter)
Signal of Paradigm Shift

When the target audience for documentation shifts from "human developers" to "AI agents," the entire developer toolchain needs to be rewritten. This represents a trillion-dollar rebuilding opportunity for the DevTools ecosystem — from CLIs to API docs to deployment pipelines, everything must become "agent-native."

VI."Understanding Cannot Be Outsourced" — Karpathy's Warning

The most profound moment of the entire talk came during the final Q&A about education. The host asked: when intelligence becomes cheap, what is still worth studying deeply?

Karpathy cited a tweet that he "thinks about every other day":

"You can outsource your thinking, but you can't outsource your understanding."
— Cited by Andrej Karpathy

Then he elaborated:

"I'm still part of the system. Information still has to enter my brain. I feel like I'm becoming the bottleneck — just figuring out what we're building, why it's worth doing, how to direct my agents... You can't be a good commander if you don't understand what's happening yourself. LLMs really aren't good at understanding — that's still uniquely your job."
— Andrej Karpathy

Thinking vs. Understanding: A Critical Distinction

This isn't wordplay. Karpathy distinguished between two types of cognitive activity:

Thinking — Outsourceable
  • Analyzing data
  • Writing code
  • Deriving math
  • Searching for information
  • Generating solutions
Understanding — Not Outsourceable
  • Judging what's worth doing
  • Setting the right direction
  • Taste and aesthetics
  • Cross-domain intuition
  • Context for directing agents

Ghosts vs. Animals

Karpathy also discussed an essay he wrote — "We're not breeding animals, we're summoning ghosts." Animals (biological intelligence) are shaped by evolution — with curiosity, motivation, and survival instincts. AI (ghost intelligence) is shaped by data and reward functions — no intrinsic motivation, no curiosity, no empathy.

He admitted this framework "might just be philosophizing," but the core message is practical: don't treat AI like animals. Yelling at them won't make them work better or worse. They are statistical simulation circuits — pretraining is the statistical base layer, RL is the conical protrusion.

Deeper Implication

Understanding is humanity's final moat in the AI era, because understanding is not verifiable.

RL can only improve verifiable capabilities. Understanding — what is beautiful, what is worth doing, what is the right direction — has no simple verification function. This is the ultimate corollary of Karpathy's verifiability framework: the areas where AI improves fastest are precisely where humans are least needed (because AI is already good enough); the areas where AI improves slowest are precisely where humans are most needed.

VII.Historical Parallels: Who Gets Eliminated in Every Tool Revolution?

Lessons from History
1440 — The Printing Press

The Gutenberg press put scribes out of work. But it didn't replace authors — it replaced those who copied content, not those who created it. Authors' value actually soared as distribution costs plummeted.

1970s — Spreadsheets

VisiCalc and Lotus 1-2-3 eliminated massive amounts of accounting paperwork. But demand for CFOs only grew — because when calculations become instant, the value of decision-making takes center stage.

1990s — AutoCAD

Computer-aided design replaced manual drafters. Architects didn't disappear — their design capabilities were amplified 10x. What vanished were "the people who turned designers' sketches into precise blueprints."

2000s — Digital Photography

Kodak went bankrupt in 2012. Film developers and darkroom technicians all disappeared. But photographers not only survived — they thrived in a golden age powered by platforms like Instagram. What disappeared was the medium, not the eye.

2026 — Software 3.0

AI is replacing "people who write code." But like every tool revolution — what disappears is the execution layer; what remains is the understanding layer.

Historical Pattern

The formula for every tool revolution is the same: execution costs approach zero, judgment value approaches infinity. In the printing era, judgment meant "what to write"; in the spreadsheet era, it meant "what to decide after the calculation"; in the Software 3.0 era, it means "what to have AI do" and "whether what AI produces is correct and worthwhile." Karpathy's "understanding cannot be outsourced" is simply the latest expression of this historical pattern in the AI age.

VIII.Business Implications: How Companies Should Redefine "Engineer"

Implication 1: Engineering Team Size Formulas Need Recalculating

If an agentic engineer's productivity exceeds 10x, then a 5-person agentic team may equal what previously required 50-100 people. This isn't the distant future — Boris Cherny is already practicing this daily. Business leaders need to ask themselves: do I need 50 coders, or 5 agentic engineers?

Implication 2: A Revolution in Hiring Standards

Karpathy said it clearly: giving small algorithm puzzles is the old paradigm. New hiring should be: give a large project and see how the candidate orchestrates AI to complete it. The evaluation dimension shifts from "can you solve this problem" to "can you direct AI to produce a secure, efficient, maintainable solution." The core assessment is judgment, taste, and system design ability — not the ability to memorize APIs.

Implication 3: Reassessing SaaS Moats

The Menu Gen paradox is a wake-up call for every SaaS founder. Ask yourself: is my product merely glue between two AI capabilities? If AI can go directly from input to output, your application layer in between has no reason to exist. True moats are:

  • Proprietary data — data not in AI training sets
  • Proprietary verification environments — the ability to create RL environments for your vertical
  • Workflow integration — not just one-shot input/output, but embedded in enterprise processes
  • Trust and compliance — domains like healthcare, legal, and finance where accountability matters
Implication 4: Invest in "Agent-Native" Infrastructure

Karpathy's vision is clear: in the future, all documentation, APIs, and tools should be "agent-first" — designed for AI first, then for humans. This is a massive infrastructure rebuilding opportunity. Karpathy's ideal: "Give the LLM a prompt saying 'build Menu Gen,' and I don't need to touch anything — it just deploys to the web." We're still far from this ideal, but the direction is set.

Implication 5: The 18-Month Window

If Karpathy's judgment is correct — "everything can eventually be automated" — then for business leaders, the window is the period while AI is still not good at your domain. Use this time to:

  1. Build your vertical's RL environment and proprietary data
  2. Transition your team from "writing code" to "orchestrating AI"
  3. Retrofit your documentation and toolchain to be agent-native
  4. Build a trust brand — because trust is more valuable in the AI era

IX.Conclusion: Which Side Are You On?

Karpathy's talk isn't prophecy — it's description. He's not describing "what might happen in the future," but "what is already happening."

Let's review the clear picture he painted in this talk:

  1. Software 3.0 is not an incremental upgrade, but a paradigm revolution — from writing code to writing context, from commanding machines to co-creating with AI
  2. Vibe Coding raised the floor — everyone can write software now, but quality, security, and architecture still require human judgment
  3. Agentic Engineering raised the ceiling — top engineers' productivity is amplified well beyond 10x
  4. Verifiability determines automation speed — verifiable domains advance rapidly; unverifiable ones stagnate
  5. Jagged intelligence is a feature, not a bug — but it's a feature you must learn to navigate
  6. Understanding is the final moat — because understanding is not verifiable, RL cannot directly improve it

Finally, in the last few seconds before leaving the stage, Karpathy shared his view on education — tools that enhance understanding are the most interesting and exciting direction. His own LLM knowledge base project embodies this: not having AI understand for you, but having AI "project" information from different angles to give you deeper insight.

Final Verdict

In the Software 3.0 era, the winning formula has changed.

Old formula: Best engineers + best code = win
New formula: Deepest understanding + best AI orchestration ability + strongest verifiability design = win

Karpathy didn't say programmers will disappear. What he said is: "writing code" as an action will disappear, but "understanding systems" as a capability will become more valuable than ever.

You can outsource your thinking. But you can't outsource your understanding.

So what did you understand today?

X.Series Guide

Sequoia AI Ascent 2026 Deep Dive (3 parts)

Series theme: AGI is not the future — it's now, and you have an 18-month window.

  1. Part 1 (Overview): This is AGI — Sequoia's Computing Revolution Manifesto
    Sequoia partners' keynote: Computing Revolution vs. Communication Revolution, MAD strategy framework, Agent 4-stage evolution
  2. Part 2 (This article): Software 3.0 — From Vibe Coding to Agentic Engineering
    Karpathy's software revolution manifesto: LLM as computer, verifiability framework, understanding cannot be outsourced
  3. Part 3: The Great Parallel — Jim Fan's Robotics Endgame Thesis
    NVIDIA's Jim Fan: The Great Parallel theory, WAM replaces VLA, the Physical Turing Test countdown

References

  1. Andrej Karpathy, "From Vibe Coding to Agentic Engineering", Sequoia AI Ascent 2026, April 2026. YouTube
  2. Andrej Karpathy, "Software 2.0", Medium, November 2017. Link
  3. Karpathy, A., "Vibe Coding", X/Twitter, February 2025. Original tweet defining the vibe coding concept
  4. Karpathy, A., "Verifiability", personal blog, 2026. In-depth analysis of verifiability and LLM capabilities
  5. Karpathy, A., "We are summoning ghosts, not breeding animals", personal blog, 2026. AI ontology exploration
  6. Boris Cherny (Claude Code creator), Sequoia AI Ascent 2026 presentation. One engineer orchestrating 10-15 parallel AI workflows in practice
  7. Sequoia Capital, "AI Ascent 2026 Keynote: This is AGI", April 2026. YouTube
  8. Greg Brockman, "Human Attention Is the New Bottleneck", Sequoia AI Ascent 2026. The human attention bottleneck thesis
  9. Ethan Mollick, "Jagged Intelligence", research on uneven AI capabilities. Academic research on jagged intelligence
  10. Menu Gen by Karpathy — menugen.ai. Karpathy's signature vibe coding project