Video summary
Why Karpathy says it’s the “decade of agents,” not the year of agents
In this conversation with Dwarkesh Patel, Andrej Karpathy explains why he thinks AI agents will take about a decade to mature. He argues that the industry is overpredicting near-term progress, since current systems still lack key capabilities such as continual learning, strong multimodality, and reliable computer use. Karpathy also reflects on earlier AI waves, including deep learning, reinforcement learning on games, and the rise of LLMs, to show how the field has repeatedly tried to build full agents before the underlying representations were ready. He further distinguishes AI systems from animals, saying we are not building animals but “ghosts” or digital entities trained from human data.
Agents are promising, but not ready
Karpathy argues that today’s agents are impressive but still far from the reliability needed to function like an employee or intern.
What’s still missing
He points to missing pieces like multimodality, computer use, continual learning, and broader cognitive capability as major bottlenecks.
Why he expects a decade
The discussion contrasts short-term hype with Karpathy’s longer timeline, based on nearly two decades of experience in AI.
Lessons from past AI turns
Karpathy revisits major shifts in AI—from deep learning to Atari RL to LLMs—and says the field has repeatedly moved too early toward full agents.
Topics
Agent timelines
Karpathy’s case for a slower agent timeline and why “the decade of agents” is more realistic than “the year of agents.”
Current bottlenecks
The technical gaps that still prevent agents from working like useful employees or interns.
AI history and lessons
How past waves in AI shaped Karpathy’s view of what should come before full agents.
Start with the video endpoint to capture ID, channel, publish date, duration, and source context.
Pull timestamped transcript data for summarization, search, citation, and RAG preparation.
Collect visible audience comments to identify themes, objections, questions, and engagement signals.
Persist structured JSON, run analysis, and publish dashboards, alerts, or research reports.
Public transcript excerpt
Transcript
Timestamped public transcript passages group captions into readable sections, making the video easier to scan, cite, and summarize.
Show timestamped transcript excerpt(1 passage)
We have some very early agents that are extremely impressive and that I use daily—Claude and Codex and so on—but I still feel there's so much work to be done. My reaction is we'll be working with these things for a decade. They're going to get better, and it's going to be wonderful. I was just reacting to the timelines of the implication. What do you think will take a decade to accomplish? What are the bottlenecks? Actually making it work. When you're talking about an agent, or what the labs have in mind and maybe what I have in mind as well, you should think of it almost like an employee or
Related Crawlora APIs & guides
Build YouTube data workflows with Crawlora
This showcase is built from Crawlora's public YouTube data APIs. Use the same endpoints and guides to build your own transcript, comment, and creator-intelligence workflows.
More AI video examples
Browse structured transcript and comment showcases in AI.
More Podcasts video examples
Browse structured transcript and comment showcases in Podcasts.
YouTube API
Transcript, comments, and video metadata endpoints that return normalized JSON.
YouTube transcript extraction
Build searchable, RAG-ready transcript pipelines from public videos.
YouTube creator intelligence
Monitor creators, audiences, and content trends across channels.
Podcast & audio intelligence
Turn long-form audio and podcasts into structured, analyzable data.
Related showcases
More structured YouTube examples
Chip design from the bottom up – Reiner Pope
Dwarkesh Patel and Reiner Pope build AI chip design from the ground up, starting with logic gates and multiply-accumulate operations before moving into adders, precision tradeoffs, and why low-bit arithmetic is so powerful for neural nets.
What rebuilding AlphaGo teaches us about self-play, RL, and the future of LLMs
Eric Jang explains AlphaGo from the ground up, using Go’s rules, endgame scoring, and search complexity to show why deep learning made the problem tractable. The episode connects those ideas to self-play, reinforcement learning, and broader lessons for future AI systems.
How GPT, Claude, and Gemini are actually trained and served – Reiner Pope
Reiner Pope explains the mechanics behind how GPT-style models are trained and served, focusing in this excerpt on inference economics. Using a roofline-style analysis of transformer execution on a GPU cluster, he shows how batch size, weight fetches, compute throughput, and KV cache access shape latency and cost. The discussion helps explain why higher-priced fast modes can stream tokens more quickly, and why serving many users together can dramatically improve efficiency.
Audience comments snapshot
Audience reactions center on Karpathy’s charisma and the show’s draw
The sampled comments mostly focus on Karpathy’s delivery, with viewers joking about his rapid speaking and laughter, and noting the strong pull of a Karpathy-Dwarkesh conversation. A few comments also highlight the appeal of guests who are familiar with the podcast and engage with it as listeners themselves.
Comment themes
Charisma and format over substantive discussion
The public sample leans more toward lightweight, personality-driven reactions than detailed debate about the episode’s technical arguments.
Strong appeal of the Karpathy-Dwarkesh combination
The comments suggest the guest-host combination itself is a major audience draw, independent of the topic.
Audience signals
Karpathy’s speed becomes a running joke
Several commenters joke about Karpathy’s fast speech or delivery, with one saying his “tokens/sec” is breaking their context window and another quipping that it’s not running on 2x speed.
Positive reactions to Karpathy’s personality
Viewers mention Karpathy’s laugh and overall presence as enjoyable or comforting.
High anticipation for the pairing
Some comments treat a Karpathy and Dwarkesh episode as must-watch listening and say they clear their schedule for it.
Podcast credibility through guest fandom
One commenter appreciates when guests reference other episodes, framing it as a sign that the podcast has become notable enough for guests to be listeners too.
Representative public comments
Karpathy's tokens/sec is breaking my context window
I don’t know about you guys but Karpathy’s laugh always puts a smile on my face
Just realised it's not running on 2x lol
Köszönöm, uraim. While painting on canvas here in my studio in Kyiv, listening to two gentlemen who relish in learning and being knowledgeable in a very broad manner, I'm struck by two feelings: How nice it is to hear two immigrant North Americans at the forefront of social issues. And, despite all our current chall...
Am a simple man,, I see karpathy and dwarkesh i clear my schedule
I always love it when the guests casually mentioned they listen to other episodes. You know you’ve made it when not only do you have the coolest guests but those guests are actual fans.
Use Crawlora's YouTube comments API with the video and transcript endpoints to collect viewer language, thread activity, and audience signals.