Feb 16, 2023 · 1h 4m · a16z
The 1000x Developer with Amjad Masad
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
In this episode of the a16z Podcast, host Steph Smith interviews Replit founder and CEO Amjad Masad about how cloud platforms and AI tools are revolutionizing software development, drastically shortening learning curves, and creating the next generation of highly empowered 1000x developers.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The host holds 32.6% of the talking time here. How this is scored →
The host as informed peer 4.6 Guest teaching 3.9 Guest disagreement 0.8 The host pushing back 0.7
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 59:13 Attack on AI Safetyism and Government Regulation
Amjad forcefully attacks the safetyism movement, asserting that anyone talking about safety probably wants something bad for you and comparing state safety impulses to Japanese internment camps.
Hardest push from the host ▶ 49:00 Challenging Replit Bounties DifferentiationSteph directly challenges how Replit Bounties differs from existing freelance platforms like Upwork that perform vetting and matching, forcing Amjad to justify their integrated IDE model.
Biggest teaching moment ▶ 41:08 Explaining Action Models vs Large Language ModelsAmjad educates Steph on action transformers, explaining how training models directly on raw mouse and keyboard browser streams allows AI to execute web workflows without traditional APIs.
The host holds their own ▶ 36:15 Framing Emergent AI Possibilities via Chess and Flight SimSteph demonstrates deep technological domain literacy by synthesizing AlphaGo's Move 37 and Flight Simulator 3D rendering to ask whether AI coding will unlock entirely new programming modalities.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Legal Disclaimer | 2 | 2 | 0 | 0 | Steph opens with a legal disclaimer and asks Amjad to introduce Replit and explain how it differs from traditional IDEs. Amjad provides a collaborative, high-level overview of Replit's mission to reduce friction between idea and product. | |
| Early Replit Stories & Rapid Fire Game Setup | 3 | 3 | 0 | 0 | Steph introduces a structured rapid-fire guessing game testing extreme user stories. Amjad shares community metrics, gently correcting her assumption that 50 projects is high for prolific Replit users. | |
| Age Outliers, Community Connections, and Extraordinary Prodigies | 2 | 2 | 0 | 0 | Amjad shares outlier user stories, including an elderly physics researcher publishing code and a six-year-old Microsoft-certified programmer. Steph listens enthusiastically to the prodigy anecdotes. | |
| Life-Changing Impact: The COVID App Developer Story | 4 | 3 | 0 | 0 | Amjad tells the story of an Indian student building a viral COVID relief application. Steph demonstrates understanding of developer friction by sharing a soccer gear prep analogy regarding setup overhead. | |
| Ghostwriter: AI Pair Programming and Reimagining the IDE | 5 | 3 | 0 | 0 | Amjad outlines Ghostwriter's AI pair programming and CLI features. Steph draws on her own self-taught coding experience to discuss how AI makes error feedback less binary, comparing it to gradient descent. | |
| Why Replit Built In-House AI Models vs. Third-Party APIs | 6 | 4 | 0 | 2 | Steph asks a probing technical question about why Replit built in-house AI models rather than relying on third-party APIs like Copilot. Amjad details hardware-software integration, latency control, and data flywheels. | |
| Push vs. Pull Modalities & The 'Society of Models' | 6 | 4 | 0 | 0 | Steph references technical insights from Daniel Gross and Nat Friedman to frame interface trade-offs. Amjad explains push vs pull UX modalities and Replit's society of models design. | |
| The Future Developer Ecosystem: The 1000x Engineer and Bimodal Distribution | 5 | 5 | 0 | 0 | Steph frames a dual hypothesis about whether AI creates lazy developers or better engineers. Amjad details a bimodal distribution where mid-tier glue-code jobs vanish while elite developers become 1000x productive. | |
| Software Leverage and Debunking 10x Denialism | 5 | 4 | 2 | 0 | Steph builds on leverage concepts by comparing software scaling to physical trades like plumbing. Amjad sharply criticizes 10x denialism, labeling it as politically motivated discomfort with individual capability differences. | |
| Accelerated Learning: Building Real Applications in 26 Hours | 4 | 5 | 0 | 0 | Steph offers her personal dataset of tracking 300 hours to reach proficiency. Amjad reframes learning speed by describing a student who built and monetized a viral app after just 26 cumulative hours of study. | |
| Unlocking New Possibilities with LLMs and Action Models | 7 | 5 | 0 | 0 | Steph shows strong domain synthesis by referencing Move 37 in Go and Flight Simulator's 3D terrain generation to ask if AI will yield emergent programming capabilities. Amjad introduces action transformers trained on browser streams. | |
| The Rise of Mobile and Tablet Coding | 4 | 4 | 0 | 0 | Steph notes the historical rarity of mobile and tablet coding. Amjad explains how conversational prompting turns phone development into an iterative swipe-based workflow that he calls Tinder for code. | |
| Replit Bounties and the Future of Software Labor Markets | 6 | 5 | 1 | 4 | Steph challenges Amjad on how Replit Bounties genuinely differentiates from established freelance marketplaces like Upwork. Amjad defends their edge via integrated cloud execution runtimes rather than static zip files. | |
| Rapid-Fire Questions: AI's Impact on Society and Opportunity Inequality | 4 | 4 | 1 | 0 | Steph presents a series of macro AI questions inspired by Sam Altman. Amjad emphasizes how AI expands equality of opportunity while simultaneously widening economic disparity and societal envy. | |
| Neutral Technology, Open Source, and Preventing AI Monoculture | 5 | 4 | 3 | 2 | Steph posits that tools are neutral but their real-world applications are not. Amjad forcefully pushes back, arguing that tech must remain strictly neutral and open source to prevent Silicon Valley monoculture. | |
| AI Safetyism, Governance, and Optimism for Builders | 6 | 6 | 6 | 3 | Amjad aggressively attacks AI safetyism as nanny-state overreach, comparing safety mandates to Japanese internment camps. Steph pushes back by citing rapid consumer adoption dynamics and historical tech panic. |
Statements from this episode (27)
Masad: Replit users are building functional software in 30-minute increments
“And we're seeing people making software in a 30 minute increments. Someone the other day put a request for a landing page up. They attached a figma file. They got it back in 30 minutes. That's really never been done before.”
Steph Smith: Self-teaching coding required roughly 300 tracked hours
“My journey took around 300 hours. Yes, I tracked it.”
Masad: Replit Bounties enables hiring AI and human developers for projects
“If you don't think coding is for you, then you can hire someone from our community to make the program for you. So we have this thing called repli bounties where You can just like put a price on a project you want to get done, describe the project, and then a …”
Masad: A 13-year-old built a crypto mining operation on Replit
“One is recently shared a thirteen-year-old kid that set up a sizable crypto mining operation via Replit.”
Masad: Self-taught Replit user landed software engineering role at Google
“We've had someone who learned to code on Replit primarily on their own, became a very productive community member, and applied to work at Replit. And we thought they're very talented, but at the time there wasn't like a really good fit for them. And so they st…”
Masad: YC startup Fig built its initial product entirely on Replit
“FIG is this command line autocomplete tool that this YC startup, I think it was like YC-twenty or something like that, or YC-twenty-one. And it's used by tens of thousands of engineers And then the original product was entirely built on, on Replit.”
Masad: A 70+ year-old physics PhD published an academic paper using Replit
“Someone who ended up working at Replit, their grandpa in India is like a physics PhD. They do a lot of, like, physics sort of writing, and in one recent paper they published, they used Repl.it to write physics simulation, and Repl.it was published as part of t…”
Masad: Replit crashed due to viral COVID relief app built by teenager
“That kid wrote the application that first responders are using to find equipment, to find oxygen, to find all these things. And it went totally viral. It brought down our website.”
Masad: Tech companies build proprietary tech rather than glue code wrappers
“If you call yourself a technology company, that means you build technology, right? It just doesn't mean you're just like building glue code on top of like existing technology.”
Masad: Replit still relies on OpenAI for large-model AI workloads
“I should say that we still use OpenAI for a lot of the sort of like you know, bigger workloads That are quite really large models.”
Masad: Proactive AI code completion requires smaller, low-latency models
“The push model is actually fairly expensive. Because you're computing all the time. It needs to actually be fairly low latency. So that commits you to a smaller model. So you can't use a super large model.”
Masad: Replit uses three distinct AI models while Copilot uses one
“Copilot uses a single model, rapidly uses like three different models of different sizes. So the smallest models as the model, you know, over your shoulder, continuously giving you suggestions, then a medium sized model to do the transformations and things lik…”
Masad: Global developer count will reach 300 million by 2030
“So with software engineering becoming more accessible because of AI, I do think we're going to get a lot more developers. I think we're going to get way, way more developed, probably 10 X. So right now there's like thirty million developers in the world. There…”
Masad: AI will eliminate mid-level 'glue code plumber' developer jobs
“And the way I see, so the developer market evolving to accommodate this new technology Is that there's going to be this bit of a bimodal distribution. So bimodal distribution, meaning there's no middle end, there's a large tail on both sides, right? So the mid…”
Masad: AI will create 1000x engineers, shrinking core tech teams
“And so you're gonna have maybe fewer of those sort of low level, 10 X engineers, but they're gonna be a thousand X engineers. And so maybe a single company would need two or three Like elite engineers, and then maybe dozens of front end engineers kind of build…”
Amjad Masad: Denial of 10x engineers is politically motivated
“The 10 X denialism is, is kind of funny to watch. Like every one of those people that say 10 X engineers don't exist. Know in their heart of hearts that X engineers exist, and they probably worked with someone they highly respected and admire. The reason they …”
Masad: AI makes learning to code more valuable by elevating all developers
“To me, it feels like it's more worth it today because it's a tide that raises all boats, right? Like we said, the 10 X becomes a thousand X. But like a .1 X becomes a, you know, a 10 X, right? So if you're someone who's previously maybe whose impact with codin…”
Masad: Replit user built AI site and earned $1,800 after 26 hours
“So in 26 hours, they were able to do to build an entire website using AI and earn 1800 dollars from biology.”
Masad: GPT-3 slashes parser development time from weeks to 15 minutes
“GPT-III is incredibly good at parsing. If you give it a malformed JSON, it will still parse it. Writing parsers is one of the hardest things you can do in programming. Writing parsers in GPT-III is one of the easiest thing you could do. You could spend 15 minu…”
Masad: Action transformers will turn every software interface into an API
“If action transformers Became mainstream and as powerful as GPT, then I think it'll unlock a new programmable platform because now you can sort of, it's almost like everything has an API suddenly.”
Masad: Generating software by talking to phones will obviously happen
“Being able to generate software by talking to your phone It's going to be a very clear thing that will happen. At minimum, being able to like instruct your AIs while you're on the go and review their work, that's obviously going to happen.”
Masad: Full-time employment is a market bug caused by high transaction costs
“The fundamental observation is that full-time employment is a bug. It's not a feature of the market. The reason why full-time employment exists is because the transaction cost of doing something is really high.”
Masad: Software development will become a fluid AI-human labor marketplace
“I think once you add all these sort of innovations together, I think you're going to get this like super fluid market of AIs and humans that you can go from an idea to a product just like that.”
Masad: AI alignment reflects Silicon Valley sensibilities, not average humans
“I think a lot of what's called AI alignment today is not really aligning with what the average human being wants. It's aligning with like what the sort of Silicon Valley average sensibility is, which I don't think it's good.”
Masad: AI technology should be neutral, letting users decide applications
“And they hate the fact that it can be neutral and they think that neutral is bad. And I think that's where we really need to push back is that no, actually it should be neutral. And then the user have to decide what to do with it.”
Masad: Claims of safety are often used to mask harmful control
“Anyone's talking about safety, they probably want something bad for you. And so that, that would be my bias is that safety is bad or at least when they want to impose something in the, Under the guise of safety, I would be super skeptical.”
Masad: Regulating AI before full deployment will cause major trouble
“And the problem now is that there are some people talking about AI takeoff being so fast that, you know, you need to react to it really quickly. And I think that'll get us into trouble. Actually, if you're gonna force politicians To understand this technology …”