Skip to content
Y CombinatorY Combinator

Andrej Karpathy: Software Is Changing (Again)

Andrej Karpathy's keynote on June 17, 2025 at AI Startup School in San Francisco. Slides provided by Andrej: https://drive.google.com/file/d/1a0h1mkwfmV2PlekxDN8isMrDA5evc4wW/view?usp=sharing Chapters: 00:00 - Intro 01:25 - Software evolution: From 1.0 to 3.0 04:40 - Programming in English: Rise of Software 3.0 06:10 - LLMs as utilities, fabs, and operating systems 11:04 - The new LLM OS and historical computing analogies 14:39 - Psychology of LLMs: People spirits and cognitive quirks 18:22 - Designing LLM apps with partial autonomy 23:40 - The importance of human-AI collaboration loops 26:00 - Lessons from Tesla Autopilot & autonomy sliders 27:52 - The Iron Man analogy: Augmentation vs. agents 29:06 - Vibe Coding: Everyone is now a programmer 33:39 - Building for agents: Future-ready digital infrastructure 38:14 - Summary: We’re in the 1960s of LLMs — time to build Drawing on his work at Stanford, OpenAI, and Tesla, Andrej sees a shift underway. Software is changing, again. We’ve entered the era of “Software 3.0,” where natural language becomes the new programming interface and models do the rest. He explores what this shift means for developers, users, and the design of software itself— that we're not just using new tools, but building a new kind of computer. More content from Andrej: https://www.youtube.com/@AndrejKarpathy Thoughts (From Andrej Karpathy!) 0:49 - Imo fair to say that software is changing quite fundamentally again. LLMs are a new kind of computer, and you program them *in English*. Hence I think they are well deserving of a major version upgrade in terms of software. 6:06 - LLMs have properties of utilities, of fabs, and of operating systems → New LLM OS, fabbed by labs, and distributed like utilities (for now). Many historical analogies apply - imo we are computing circa ~1960s. 14:39 - LLM psychology: LLMs = "people spirits", stochastic simulations of people, where the simulator is an autoregressive Transformer. Since they are trained on human data, they have a kind of emergent psychology, and are simultaneously superhuman in some ways, but also fallible in many others. Given this, how do we productively work with them hand in hand? Switching gears to opportunities... 18:16 - LLMs are "people spirits" → can build partially autonomous products. 29:05 - LLMs are programmed in English → make software highly accessible! (yes, vibe coding) 33:36 - LLMs are new primary consumer/manipulator of digital information (adding to GUIs/humans and APIs/programs) → Build for agents! Some of the links: - Software 2.0 blog post from 2017 https://karpathy.medium.com/software-2-0-a64152b37c35 - How LLMs flip the script on technology diffusion https://karpathy.bearblog.dev/power-to-the-people/ - Vibe coding MenuGen (retrospective) https://karpathy.bearblog.dev/vibe-coding-menugen/ Apply to Y Combinator: https://ycombinator.com/apply Work at a startup: https://workatastartup.com

Jun 19, 202539mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

LLMs reshape software: English programming, autonomy sliders, and agent-ready infrastructure

  1. Karpathy frames software evolution as Software 1.0 (hand-written code), Software 2.0 (neural network weights trained from data), and Software 3.0 (LLMs programmed via English prompts).
  2. He argues LLMs function simultaneously like utilities (metered APIs), fabs (high CapEx, centralized know-how), and especially operating systems (a platform ecosystem with “apps,” memory, and orchestration).
  3. Because LLMs are “people spirits” trained on human text, they have emergent human-like psychology: superhuman recall and synthesis alongside hallucinations, jagged intelligence, amnesia-like context limits, and security gullibility.
  4. The most practical near-term opportunity is “partial autonomy” products that optimize a human-in-the-loop generation→verification loop using good UX, context management, multi-model orchestration, and an adjustable autonomy slider.
  5. LLMs make “everyone a programmer” (vibe coding), but real-world shipping is constrained by deployment/auth/payments/DevOps friction—driving a need to redesign digital infrastructure and documentation for agents as first-class users.

IDEAS WORTH REMEMBERING

5 ideas

Treat prompts as a real programming layer (Software 3.0).

LLMs are programmable computers where English is the interface, so product teams should deliberately choose among 1.0 code, 2.0 training, and 3.0 prompting depending on reliability, cost, and control needs.

Design LLMs into products as a platform (an ‘LLM OS’), not a chat box.

Successful apps (e.g., Cursor, Perplexity) wrap the base model with context management, multi-call orchestration, and task-specific UI so users don’t have to “talk to the OS through a terminal” all day.

Optimize the human-AI generation→verification loop; verification is the bottleneck.

Since humans must audit fallible outputs, interfaces should make checking fast (diffs, citations, review controls) and keep changes small enough to validate rather than dumping massive, unsafe outputs.

Use an autonomy slider instead of betting everything on fully autonomous agents.

Offer levels from assistive suggestions to file-level edits to repo-wide actions, letting users adjust autonomy to task risk and complexity while the underlying systems mature.

Account for ‘LLM psychology’ like you would with a quirky but capable coworker.

LLMs can be encyclopedic yet hallucinate, fail on simple edge cases (jagged intelligence), forget across sessions (context limits), and be vulnerable to prompt injection—so products need guardrails and supervision mechanisms.

WORDS WORTH SAVING

5 quotes

I think fundamentally the reason for that is that, um, software is changing, uh, again.

Andrej Karpathy

Neural networks became programmable with la- large language models. And so I, I see this as quite new, unique. It's a new kind of a computer.

Andrej Karpathy

Basically, your prompts are now programs that program the LLM. And, uh, remarkably, uh, these, uh, prompts are written in English, so it's kind of a very interesting programming language.

Andrej Karpathy

I think it's kind of fascinating to me that when the state-of-the-art LLMs go down, it's actually kind of like an intelligence brownout in the world.

Andrej Karpathy

So the way I like to think about LLMs is that they're kind of like people spirits. Um, they are stochastic simulations of people.

Andrej Karpathy

Software 1.0 vs 2.0 vs 3.0 framingProgramming LLMs in English (prompts as programs)LLMs as utilities, fabs, and operating systems1960s time-sharing analogy and early platform formationLLM psychology: superpowers, hallucinations, jagged intelligencePartial autonomy apps, GUIs, and autonomy slidersBuilding docs/infrastructure for agents (Markdown, MCP, agent-readable interfaces)

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.