Replit The Future Is Actually Very Human

Over the past few months, I’ve spent a significant amount of time using Replit, and after successfully shipping multiple apps with it, I’ve had quite a few people ask for my thoughts. So, here we are. Edit: Replit released v2.0 of their agent about one month ago (early access). It has shifted many of my thoughts which I note below. Broadly speaking, I feel strongly Replit is the best “prompt to code” tool aiming at less technical folks (or completely non-technical folks).

Replit is a rocketship of a startup right now, and for good reason. Edit #2 6/30: I’ve moved a lot of my workflow to Cursor and am using Replit less but still find it an awesome product, and don’t plan on cancelling my subscription. If you want my bull/bear case on the company as of now, check this (recent) post.

A Quick Background on Replit For those unfamiliar, Replit was founded in 2016 as a simple, browser-based coding environment that allowed users to experiment with different programming languages—eliminating the hassle of setting up local environments. The company was started by Amjad Masad, who, while growing up in Jordan, struggled to access coding environments. That challenge inspired him to build a browser-based solution, making coding more accessible to everyone. (Side note: If you’re even remotely interested in computer science, I highly recommend checking out Amjad’s blog.

It’s insightful, well-written, and packed with interesting takes.) What Replit Has Become Fast forward to today, and Replit has evolved into a powerful cloud-based IDE, integrating agentic AI coding platform and automated deployment pipelines. So, what does that actually mean? It means you can log into Replit, interact with an AI agent, and build end-to-end software in multiple programming languages—without necessarily writing code yourself. The platform includes: - An “agent”, which lets you interact with your codebase via prompts, effectively coding through natural language.

An “assistant”, similar to the agent but designed to handle smaller tasks per prompt (and at a lower cost). - Git integration, so you can version control your projects seamlessly. - A command-line interface (CLI) for direct terminal access. - Secrets and dependency management, making it easy to handle API keys and project requirements. - A fully functional deployment pipeline, so you can take a project from development to production in a streamlined workflow. - Collaboration tools, enabling teams to work on projects together.

Shell, so you can access your app direct through command line interface - Of course, all the essentials—code editing, searching, and execution—right in your browser. - …and much more My Experience with Replit For some context, I don’t have a traditional programming background. While I’ve taken a handful of online courses , I am not a software engineer (at all). That said, I’ve shipped a handful of little stand alone scripts, macros and “programs” (in quotes for a reason) that solved real problems.

My real expertise and career has been on the product and operations side of technology businesses, where I work closely with teams of engineers, designers, infrastructure personnel and my curiosity has served me well with respect to gaining a broad understanding of “how software works”. I’ve written countless PRDs and my systems design/architectural understanding is strong. I can sketch out how most any startup app might work, and love geeking out on the ByteByteGo schematics showcasing how bigger systems work.

It turns out that this conceptual grasp of software architecture is very important when using any vibe coding platform effectively. One thing I can’t overemphasize enough is how important it is for you to be able to write requirements clearly and concisely. Any good product manager understands this, but when working with an AI, with limited context windows, this is even more important. You need to “chunk” your project, even simple MVPs, into various phases to become successful.

The order you approach your app build, and how much you feed “the robot” all at once is often the difference between success and failure. Since December, I’ve built ~5 different apps on Replit, including: - A dating app - An HOA payment portal (pulled offline; only spinning it up once per year for payments) - An admin panel for that dating app - Tucker and the Robots: MP3 enabled “album launch” website for an AI band I spun up. This was built in minutes.

Real Estate Analysis Tool: Upload your MLS CSV and this will analyze the market for you. - An SEC filing aggregator; currently using this to fine tune Deepseek V3 on financial filings. - A computer vision app for Rali. (stuck on tweaking the computer vision model dangit!) - Various landing pages (yes, they all look the same at a point, but its a great way to ship something fast) - Worth noting, I’ve also failed building a number of apps too.

(3x as many; for a running list of apps in production, go here https://jeffbrines.com/projects-apps/ ) All in I’ve spent roughly $361 on subscriptions and additional usage since early December. Considering how much I’ve learned and what I’ve been able to test, I consider this a bargain. Here’s what I’ve learned: Key Takeaways 🚀 Replit is an incredible teaching tool The ability to “chat” with my codebase makes learning far more dynamic than traditional programming environments.

Being able to ask questions and see real-time results has dramatically improved my ability to understand coding concepts. I learn best this way; I didn’t learn how to wrench on snowmobiles by first studying thermodynamics, I learned by diving into a completed snowmobile and working from there. This is how it feels to me working with Replit. 🧠 You need to be somewhat technical The better your understanding of systems and architecture, the better your results with Replit.

You don’t need to be a top-tier software architect, but curiosity, a baseline understanding of software design, and the ability to work in other LLMs to unpack problems combined with a willingness to grind through issues will serve you well. ⚡ The agent makes setup frictionless Replit’s AI agent does a great job setting up the development environment and installing dependencies automatically, letting beginners get started way faster.

If you’ve ever tried to setup your own development environment for the first time, you know this can really get in the way of the fun parts of programming. Replit makes this seamless and easy. I can’t stress how important this really is. 🛠️ Debugging can be hair pulling (but this is true with or without an AI agent) While agentic debugging works well, you still need to understand how to use dev tools to guide the agent effectively. Update; this is *vastly* improved in their agent 2.0. Like…10x better.

✍️ Prompt writing is very important The size and quality of your prompt matter a lot. You need to write precise, succinct prompts that chunk work into properly scoped user stories—not too big, not too small. This is where my background in product really helps. 📈 The bigger your codebase, the more the agent struggles This is expected, given how context windows work in large language models (LLMs). AI can only “remember” so many tokens at once, which is a limitation for all AI-driven coding tools.

I expect improvements here over time as KV-cache gets bigger on the hardware side and we have breakthroughs architecturally on the model side. 🔄 Git is your best friend If you don’t understand Git, you’ll learn fast—or quit. The AI, like a human developer, will make mistakes and regress. Knowing how to revert changes is critical.

(On a related note, Replit integrates perfectly with GitHub, making it easy to port your projects outside of the Replit environment.) 🏗️ Integrating third-party APIs is seamless I’ve used AWS S3, Stripe, OpenAI, and Polygon without issues. The agent handles secrets securely, ensuring API keys aren’t exposed inappropriately. ⚔️ Sometimes, you fight the agent There have been cases where I set up my dev environment correctly, only for the agent to make changes that break it.

Unfortunately, you can’t prevent the agent from modifying certain files or sections of your codebase yet. edit; this has been fixed in agent v2.0 for the most part. 🔥 Dev vs. Prod: A big flaw You work off production data by default, which is a major no-no in software development. Replit doesn’t make it easy (one click) to separate dev and prod environments. This is something I’d love to see them tackle, head on, in the near future.

Worth noting, its not hard to setup multiple databases if need be, but it could be much easier. 🔍 Sometimes, GPT-4 or Claude is better for debugging When stuck on a tricky bug, I’ve often used GPT o1 or Claude alongside Replit’s AI tools. At times I’m asking one LLM what a function does or why we are using a particular library while revising a prompt in another LLM. The combination has made me more effective than I ever thought possible as a non-traditional programmer.

edit; this is less true with v2.0. 🎨 The designs all look the same This isn’t surprising, as the agent leans on popular frameworks and libraries. But don’t be shocked when your app looks just like your friend’s. Who is Replit For? In its current form, Replit’s product-market fit (PMF) seems limited. - Experienced software engineers will likely stick to Cursor. - Non-technical users might find more success with tools like Lovable, Bolt, or n8n (for workflow automation).

Obviously, you can still have success with Replit, but you will go one of a handful of ways ways – one shot the project and iterate briefly to success, hit a wall, get frustrated, hit your $25 limit and give up, or get curious (and semi-addicted) to the process, keep working until you have something. edit; with v2.0 I completely change my mind on this; Either go with Cursor if you are a pro or Replit if you are not a pro; simple as that.

Replit excels at quickly prototyping MVPs and testing ideas against a market. And I strongly believe that speed and iteration are key factors in software success. But if you see strong traction, you’ll likely need a real engineer to take your product from 0 to 1. The Big Question: How Many People Really Want to Build Software? This is something Replit is wrestling with.

If the total addressable market (TAM) for software development is much bigger than it currently appears—because tools have historically been too complex—then Replit is on its way to becoming a multi-billion-dollar company. However, if software development doesn’t democratize the way video content creation did in the mobile era, then Replit needs to win over professional engineers to stay competitive. That means building tools that rival Cursor. The irony? Replit has more raw technological firepower than Cursor, but it’s aimed at a different audience.

If they can create a product that serious engineers love too? Watch out. (And hey, Replit—hire me.) 2 Comments Comments are closed.

People Also Asked

• Replit — The Future is Actually Very Human
• The Future is Actually Very Human - nontrivial.ai
• My Thoughts on Replit After Several Months of Use - Jeff Brines
• Gap Between Average Replit User And Senior Google Engineer Will Vanish ...
• OmniEleven 万千十一
• ReplitKeeps Breaking My Nuxt Code: Here's Why (And How to Fix It)
• Las 7 Mejores Alternativas aReplitpara Lanzar Aplicaciones Rápido...