Mar 11, 2026 · 24m · startup-ideas
Karpathy's "autoresearch" broke the internet
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Greg Isenberg breaks down Andre Karpathy's viral open-source tool AutoResearch, explaining its autonomous experimentation mechanics, ten actionable startup and business use cases, and how to deploy it using cloud GPUs.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Greg holds 99.9% of the talking time here. How this is scored →
Greg as informed peer 0.0 Guest teaching 0.0 Guest disagreement 0.0 Greg pushing back 0.0
speaking balance: gold is Greg, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 8:34 Brief automated or producer verbal nod
In an entirely solo presentation, the only guest presence is a passive one-word acknowledgment, with zero real combativeness throughout.
Hardest push from Greg ▶ 12:35 Host counterpoints blind trust in AI tradingGreg self-corrects the hype around autonomous quant trading, emphasizing that users will lose money without human-in-the-loop oversight.
Biggest teaching moment ▶ 16:50 Highlighting Agent Hub architectural conceptsGreg reads Karpathy's and Morgan Linton's posts explaining advanced concepts like bare git message boards for agent swarms.
Greg holds their own ▶ 20:20 Practical GPU workaround walkthroughGreg demonstrates hands-on practical expertise by guiding non-NVIDIA local users through using Google Colab and Claude Code to deploy the repo.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Greg as informed peer | Guest teaching | Guest disagreement | Greg pushing back | Why |
|---|---|---|---|---|---|---|
| Core Architecture and Mental Model of AutoResearch | 0 | 0 | 0 | 0 | Greg delivers a solo monologue breaking down Andre Karpathy's AutoResearch architecture, explaining the GPU training loops and mental models. Per instructions, monologue segments receive zeroes across all dynamic scoring categories. | |
| Announcement: Building Businesses in the Age of AI Event | 0 | 0 | 0 | 0 | Greg makes an announcement for an upcoming live event on building businesses in the age of AI, followed by A/B testing business concepts. This is an announcement/monologue segment scored at zero. | |
| Business Use Cases: Research Services, In-App Tools, and Testing Agencies | 0 | 0 | 0 | 0 | Greg outlines commercial applications including research-as-a-service, native SaaS optimization buttons, and testing agencies. The monologue continues without interactive debate. | |
| Business Use Cases: Auto-Quant Trading, Lead Qualification, and Finance Ops | 0 | 0 | 0 | 0 | Greg explores auto-quant trading rules, CRM lead qualification, and AP finance ops automation. This is an uninterrupted solo ideation segment. | |
| Business Use Cases: Internal Productivity Labs and Living Due Diligence Shops | 0 | 0 | 0 | 0 | Greg shares startup ideas for internal productivity labs and living due diligence reports, referencing Morgan Linton's clinical trial ideas and Karpathy's Agent Hub. Monologue scoring rules apply. | |
| Technical Walkthrough: Setting Up AutoResearch on Cloud GPUs | 0 | 0 | 0 | 0 | Greg walks through the technical setup of AutoResearch using Claude Code and Google Colab with cloud GPUs. This is an instructional solo walkthrough. | |
| Reflections on Solo Podcasting and Final Thoughts | 0 | 0 | 0 | 0 | Greg reflects on audience feedback regarding solo podcast episodes and wraps up with final thoughts and channel sign-offs. |
Statements from this episode (8)
AutoResearch requires an NVIDIA chip or cloud GPU, excludes M1 Macs
“You need a NVIDIA chip to actually run auto research, or you can do it in the cloud. I'll talk about this at the end of the episode, but you know, you do need that. You can't just run it on, let's say you have a Mac book M one or something like that.”
Autonomous AI loops will replace A/B testing tools like Optimizely
“Think of, you know, tools like Optimizely. That's a SaaS tool that, you know, when I first moved to San Francisco, I remember how big they were and everyone's talking about Optimizely and, A B testing. And it's like, well, this is the future of that auto resea…”
SaaS products can embed AutoResearch loops as enterprise tier upsells
“So if you already have built a SaaS or workflow, embed an auto research style agent so your users can press optimize, just like a big, I envision like a big button that just says optimize, and the system runs a mini research loop for them. So for example, tune…”
AutoResearch lets agencies offer 100x more testing for the same fee
“Because Autoresearch lets you run hundreds of experiments instead of a few, you have a simple pitch. We do a hundred times more testing than other shops for the same or lower fee.”
Autonomous agents will empower retail traders to bypass traditional financial advisors
“It's going to be an unfair advantage for a lot of people. So I think you're going to see a lot more digital products that people sell and also, you know, just using their own money, trading themselves instead of giving one percent or whatever to a financial ad…”
Karpathy launched Agent Hub, a GitHub equivalent designed for AI agents
“Karpathy also launched agent hub. What is agent hub? It's GitHub for humans. Sorry. GitHub is for humans. Agent hub is for agents. So it's basically a GitHub for agents and agents form collaboration platform, a very promising direction.”
Agent Hub abandons main branches and pull requests for AI swarms
“Agent first collaboration platform, a bare git repo, a message board designed for a swarm of agents working on the same code base. Think of it like a stripped down GitHub. Where there's no main branches, no main branch, no PRs, no merges, a sprawling DAG of co…”