This AI Tool Lets You Build Apps Faster Than Googling | Amjad Masad (Replit)

Why personal software is the future of coding and a live demo of Replit's AI agent. Plus, how to find job security in tech in the AI era

Dear subscribers,

Today, I want to share a new episode with Amjad Masad.

Amjad is the CEO of Replit, an AI coding platform with 34M users that has made it possible for anyone to build an app in 60 seconds from their phone. In our chat, Amjad shared how you can learn to code with AI and gave me a live demo of Replit’s AI agent. We also discussed the skills that will still matter in AI.

Watch now on YouTube, Apple, and Spotify.

This episode is brought to you by Vanta — Join over 9,000 global companies like Atlassian, Quora, and Factory who use Vanta to manage risk and prove security in real-time. Get $1000 off at vanta.com/peter

Amjad and I talked about:

• (00:00) Two paths to tech job security in the AI era

• (06:36) The best way to learn to code with AI

• (07:41) Building a nutrition tracking app with Replit

• (14:07) How to avoid getting stuck on bugs with AI

• (18:30) 5 tips to get the most out of AI coding agents

• (22:10) Why everyone should build personal software

• (24:00) What makes Replit's AI approach different

• (27:41) Why traditional roadmaps are dead

• (30:40) How Replit uses AI internally to build faster

• (34:00) Build an app in 60 seconds from your phone

• (36:14) Career advice for tech folks to become AI native

---

How to learn to code in the AI era

Get the Replit iOS or Android app to make apps in 60 seconds with your phone

Welcome Amjad! Let’s start by talking about how to learn to code with AI. Jensen famously said that kids shouldn’t learn to code because AI will just do all the work. But you have a law that says otherwise, can you talk about that?

My view is that programming is evolving. When Jensen makes statements like that, I think he means that we should adapt how programming is taught.

Programming with AI is very different from the traditional approach of breaking down problems into components, writing classes, and figuring out packages. On Replit, we automate many of these chores.

What's left is you and your creativity — which is the most exciting part of programming. So my law is that the return on learning to code doubles every 6 months.

The law is based on my observation that every six months we get a new AI model that's better at coding. This pattern started with GPT-3 and we’ve seen huge jumps with GPT-4, Claude, and now new reasoning models like O3 Mini and R1.

So basically, every six months, coding becomes both more accessible for beginners and more powerful for experts. While some skills might become outdated, the coding skills you learn will take you further every six months.

So I’ve tried and failed to learn to code a few times. I even tried Replit’s 100 days of coding course but only got to like day 7. So let’s make this really practical – why is learning to code hard and how should I start today with AI?

Start with an AI coding like Replit Agent. Let’s do a live demo, what do you want to build?

Let’s build a basic nutrition tracker. I have problems tracking my protein intake and MyFitnessPal sucks.

Replit’s “improve prompt” feature kind of automates what PMs do 😅

Great, so let’s just ask it to build a nutrition tracker. If I hit “Improve prompt” it’ll write a full spec from that single line.

Ok you just automated what PMs are supposed to do.

Haha, yeah. Now you see the agent constructing a plan with the prompt. It's suggesting Streamlit because it inferred this would be a data and visualization heavy application. You can change that – I could specify React instead and it would propose a new plan. But let's go with the recommended solution.

The reason to start this way is you immediately understand what coding is for.

Traditional teaching starts with programming concepts like if statements, but you don't know what to do with them. Here, we invert that by showing what you can build.

How do you avoid getting stuck in a loop where you're constantly asking AI to fix bugs without understanding how the code works?

There are a few ways:

1. Read the comments. The AI is explaining what it’s doing as it codes.

2. Highlight code and ask for an explanation. e.g., I can highlight database.py and click "Explain with Assistant" to understand it step by step.

3. Ask higher level questions. e.g., I can ask "Explain how this project is structured." The AI will break down the components – it's a nutrition tracking application built with Python, Streamlit, and SQL Alchemy.

4. Ask AI to fix the bug and explain how. Share the bug with the Assistant and ask it to show you how to fix it.

Finally, if the AI gets stuck, use good old Google. Even before AI, engineers spent half their time on Google – it still has amazing content for learning.

My experience teaching programming and seeing people use Replit Agent is that you level up naturally, almost by osmosis.

You're exercising your debugging and systems thinking muscles. Eventually, you might become so interested in the craft that you'll take a programming course, but only after understanding what programming is for.

So it’s like learning in reverse — you first build something then break it down. I think the key step most people miss is asking questions like "explain this to me."

Yes, knowing what to ask is an important skill.

Another crucial skill is systems thinking – being able to break down problems and think through them one by one. This also helps with effective prompting.

Many people struggle with agent tools because they continuously give large-scope requests. When you tell it to do something really big each time, you increase the chance of failure versus breaking requests down into chunks.

What are your top tips to get the most out of Replit Agent?

1. Start with an AI chat session to plan out your prototype. What are the basic features? Do you need a database?

2. Create a new chat session for each feature. Update the app step by step. For example, if you want to add food lookup from an API, start fresh.

3. Use the rollback feature if something doesn't work – you can start over without breaking your application.

4. As the project grows, give increasingly specific instructions and tell it what not to do. The models can get too creative sometimes.

Also, use the assistant feature – think of it as the agent's cousin. While the agent makes decisions on your behalf, the assistant is more like a co-pilot. It won't make changes until you ask and is better for chatting and brainstorming. It's important to know when to use which tool.

---

Inside how Replit builds AI products

Thanks for giving me that demo. Let's switch gears and talk about how Replit builds products. What's your vision for the company?

Think about the ultimate function of a computer – it's meant to do things on your behalf, to automate your work. It's not just a tool like a pencil. Take an Excel spreadsheet: you change a number, and it automatically recalculates everything.

The problem is very few people know how to code. For most people, Excel is the most programmable environment they have. While there are modern tools, they don't go much beyond Excel's capabilities. If everyone could program, they'd get much more value from computers. Productivity would soar, and people would be happier not doing menial tasks.

As long as you have a machine that's good at writing programs, you have a very powerful tool. Our view isn't that everyone needs to be a programmer, but everyone can benefit from making software.

We see examples in our community – a dad makes educational software like an interactive solar system application where you can explore planets and their information. Another parent created "Chore Hero," a leaderboard showing which kids are doing the most chores. People use Replit for both personal and work projects.

This dual nature is why I believe we'll reach a billion programmers eventually. If you can use a computer, you should be able to make software.

I love how it's led to more personal software. I made an app to resize images for my blog — it addresses one person's problem.

Yes, especially since most public tools are full of ads and spam. Making your own tools is much better. Before we launched the agent, I said the ultimate test for a code generation system is that you can make an app faster than you can Google for it. In some cases, that's true today.

How do you differentiate Replit from other coding agents like V0 or Cursor?

Those apps focus on code generation, but you're not usually using them to make end-user software because you have to take it somewhere else. Sometimes people make something with V0, improve it in Cursor, then deploy it on Vercel.

With Replit, the entire software development lifecycle is in one place.

If you're using Cursor, you still need to figure out database choices, authentication, and how to set up a development environment. With Replit, you just enter a natural language prompt – we'll create a database, determine the deployment needs, handle migrations, and manage integrations. For example, we could ask Replit to summarize daily nutrition progress using OpenAI and it would handle the API integration.

Replit also supports many different stacks. People use it for:

• Data analysis and dashboards with Streamlit

• Personal apps with Flask and JavaScript or React

• APIs with FastAPI or Flask for integration with tools like Zapier

• Full-stack apps

It's a more dynamic tool with a wider range of uses.

I think most people really care about the UX of their apps. How do you get the Agent to create good UI?

Alot comes down to prompts – literally telling it to make beautiful, modern, production-ready websites. Using the right libraries is also crucial. ShadCN, a React library that V0 also uses, ships with components that look good by default.

Using an existing design system also makes apps production-ready. Lastly, it's about ensuring responsiveness, using the right HTML tags, and iterating on small details to enhance the user experience.

Let's talk about planning. You worked at Facebook for a few years where they have annual planning processes with PMs. How do you approach planning at Replit?

I’ll be honest with you.

I think roadmaps are dead.

You need to be able to react to new technology, which emerges every two weeks now. For example, we dropped everything to evaluate DeepSeek R1, determining it was academically interesting but doesn't match Claude for coding. Roadmaps can make people rigid and single-minded when the ground shifts beneath your feet.

Instead, I set high-level quarterly goals like "focusing on growth." Teams generate ideas, we pick the best ones, and execute. If working on reliability, we target specific agent metrics. We focus on 2-3 priorities per quarter, staying flexible as user behavior changes and technology evolves.

How do product teams use Replit to build Replit?

We see Replit as a tool for making standalone applications that are relatively simple to deploy and scale. For example:

• Designers use it to prototype, often skipping Figma

• Engineers test feasibility and build debugging tools

• We occasionally build production systems, typically microservices or APIs

• We use it extensively for data analysis

Do you have PMs at Replit? What do they do?

We have two PMs – one who was originally a designer and another is brand new. Our PM organization isn't fully developed yet. Our PMs are creators who:

• Build things in Figma or Replit

• Design experiments

• Talk to users extensively

• Conduct research and user testing

• Handle project management and team alignment

We're actually hiring for product management right now.

---

Make an app from your phone in 60 seconds

Let's wrap up by talking about your big launch, what’s new with Replit?

The biggest change is making Replit Agent free to try. Users get a generous credit allowance to go from prompt to working app that they can share with friends. I think sharing software you've made is one of the most cherished experiences.

We're also launching a mobile app with a novel design approach for coding on small screens. We're introducing a social experience where people can share and remix apps — that's how I learned as a kid, downloading Visual Basic code and tinkering with it.

Finally, we're rolling out a new onboarding experience. Instead of just dropping users on the homepage, it will guide them through building their first application.

---

How to find job security in the AI agent era

People have said that 2025 is the year of AI agents. How do you think this will change the structure of work?

I think if you're in school or want to work in tech, there are two divergent paths if you want job security:

1. One path is to get as close to the metal as possible. NASA won't use GPT-generated JavaScript to run rockets, and neither will Tesla for their cars. If you work on embedded systems using C, you'll have job security for maybe 50 years. Similarly, if you work on cloud platforms using Rust or Go, you'll find AI isn't very good at that because it's not well represented in training data.

2. The other path is to become a product person — a generalist, designer, and maker. The full-stack developer role is most at risk because it's heavily represented on GitHub, it's what AI is best at, and it's already being automated through tools like Replit and Cursor. Companies will increasingly look for generalists who can go from ideation to production code using AI because they'll be more highly leveraged.

If you're already in the industry, run in one of these directions. Don't stay in the middle – it's going to get squeezed.

Thank you so much Amjad! If you enjoyed this interview, follow Amjad on X/Twitter and start building apps on Replit web, iOS, and Android.