Oct 2, 2025 · 55m · latent-space
The antidote to AI fatigue — Answer.ai Solveit
Jeremy Howard · 20m spoken Eric Ries · 17m spoken Alessio Fanelli · 6m spoken Jono Whitaker · 5m spoken
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
Answer.AI co-founders Jeremy Howard, Eric Ries, and Jono Whitaker introduce Solveit, a human-in-the-loop computing environment designed to counter AI fatigue by enabling just-in-time software creation inside persistent Linux containers. Through practical demonstrations spanning semantic search, book editing, and technical publishing, they present a vision of AI that prioritizes human agency, rapid feedback loops, and general-purpose computing.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 12.1% of the talking time here. How this is scored →
The hosts as informed peer 4.9 Guest teaching 4.3 Guest disagreement 1.4 The hosts pushing back 0.9
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 23:15 Eric's warning against AI slop
Eric forcefully dismisses the idea of letting AI write prose autonomously, warning writers that unguided generation produces slop and creates cognitive complacency.
Hardest push from the hosts ▶ 22:13 Alessio queries the viability of dedicated SaaS vs personal toolsAlessio presses the team on whether Solveit's broad positioning renders traditional vertical SaaS applications obsolete for individual power users.
Biggest teaching moment ▶ 16:07 Jeremy on autoregressive error cascadesJeremy explains the underlying mechanics of autoregressive models, proving why correcting errors in-line corrupts future responses and why mutable message histories are necessary.
The host holds their own ▶ 5:38 Alessio connects latent space theory across software paradigmsAlessio articulates a technical framework linking model training as information compression to tool-mediated prompt execution as decompression, drawing on insights from Dylan Field.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome and the Origins of Answer.AI | 5 | 3 | 1 | 1 | Alessio sets up the conversation by recalling their term sheet negotiations and asks the team to explain why they founded Answer.AI and built Solveit. Eric and Jeremy explain their Edison Lab inspiration, the self-imposed 12-person constraint, and their full in-house software and services stack. | |
| Iterative Philosophy and Returning to General-Purpose Computing | 6 | 4 | 2 | 1 | Alessio synthesizes insights from his Dylan Field interview, framing AI development as expanding value from the compressed latent space. Jeremy and Eric build on this by detailing how Solveit embraces Lean Startup, OODA loops, and 1990s-style general purpose computing via persistent Docker containers. | |
| Live Demonstration of Solveit Interactive Dialogues | 4 | 6 | 2 | 1 | Jeremy demos Solveit's dialogues and live Docker environment, explaining that editing previous assistant turns prevents autoregressive compounding of errors. Jono reinforces how standard chat apps pollute context, while Jeremy shows learning mode and interactive REPL execution. | |
| Just-In-Time Software and Jono's Scent Search Demo | 5 | 5 | 2 | 1 | Alessio asks if just-in-time custom software threatens traditional SaaS applications. Jono demonstrates a perfume search engine built in an hour, while Jeremy contrasts their step-by-step human understanding approach with blind vibe coding. | |
| Eric Ries Demonstrates Book Drafting and Fact-Checking | 5 | 5 | 2 | 1 | Eric shares how he drafts his upcoming book and processes thousands of reader comments using custom Solveit dialogues while avoiding AI cliches and hallucinated fact-checks. Alessio observes that non-code workflows reveal the true power of granular context control. | |
| Dialogue Composability and Karpathy Tokenizer Video Case Study | 5 | 5 | 1 | 1 | Alessio asks how dialogues can be shared and composed like libraries. Jeremy shows an end-to-end case study turning Andrej Karpathy's tokenizer video into an interactive markdown blog post via structured dialogue engineering. | |
| Upcoming Solveit Course and Public Benefit Mission | 4 | 2 | 0 | 0 | The panel discusses the upcoming Solveit cohort, course curriculum, and their public benefit corporation mission focused on human flourishing rather than pure profit maximization. |
Statements from this episode (13)
Ries: Answer.AI will never need more than 12 employees in total
“I can remember where I was when Jeremy told me, by the way, he doesn't think the company will ever need more than 12 employees. And I was like, oh sure, you mean like this month, you know, but it's like, no, over its entire life.”
Howard: Answer.AI runs fully in-house stack without AWS or Google Cloud
“This group of, which has averaged about 10 to 12 people, currently nine, I think, have built a pretty Transformational and complex piece of software, which we can do a quick demo of later if you're interested. Using a complete web application development platf…”
Howard: Answer.AI built its own AI-powered legal and accounting firm
“Built our own fully integrated professional services firm providing legal accounting, bookkeeping. They're all using Solvert, by the way. They're all using our AI. And we were their first client, so all of our back office and everything is being done by a comp…”
Howard: Answer.ai makes adding LLM tools easier than any existing system
“We've made it easier to add tools to the LLM than anything exists in the world. Literally any Python function that you put right there in that system is immediately a tool”
Howard: Solveit provisions a persistent Linux container for every chat
“Behind the scenes, we've actually launched a persistent Linux Docker container for you and attached it to a unique URL for you. And so you actually now have your own computer. It's basically a VPS that you're now running this in. And so you can install softwar…”
Howard: Correcting LLM errors in chat history degrades subsequent model answers
“The autoregressive nature of language models means that if they make a mistake, and you correct it, and then say, no, that was a mistake, please do it this way instead. The more often you do that, the worse the dialogue answers get. Because it's in the trainin…”
Howard: Effective AI tooling requires symmetric visibility between user and model
“A key fundamental Thing of, like, basically all the AI work we do now is we realized that AI must be able to see everything that we can see exactly like we see it, and vice versa. We need to be able to see everything that the AI can see.”
Jeremy Howard: AI coding should prioritize human learning over vibe coding
“We want to make sure at the end of building something, two things have happened. The first is you understand every line of code fully, why it's there, what it does, and that you've learned from the process. So at the end of this process, you're a better develo…”
Eric Ries: AI systems cannot reliably fact-check long documents
“Fact checking a long document is still well outside the capabilities of most AI systems because, you know, for the attention reasons, but if you're working in sections, it's really easy to be like, while I'm here,”
Eric Ries reveals upcoming unannounced book titled The Incorruptible
“I'm working on a new book. It's not announced yet. So don't look too, too closely what this is, but the book is called the incorruptible. It'll come out next year.”
Ries: Vibe coding lulls users into letting AI think for them
“That really is like, to me, the number one thing I've learned from this and all forms of vibe coding and all the things that I've played with is that it's just extremely easy to get into this kind of like soporific state where the AI is doing the thinking for …”
Howard: Answer.ai is building an API to convert video discussions into text
“We're kind of building an API around this. We want to make this available, not just in Solvert, but for anybody to turn their discussions, their videos, their interviews into very, very high quality text outputs.”
Howard: Longer dialogues improve AI outputs when humans edit intermediate results
“The nice thing is that with the dialogue engineering we discussed, the longer your dialogue is, the better the AI gets, which is the opposite to what we're used to, right? Because you can edit the outputs that aren't great.”