Another AI coding tool? (Yes… but here’s why I’m building it)
I left my “safe” job about a year ago. Since then I’ve been taking on some client work to pay the bills, and on the side building coding tools to speed up my own dev process.
Like many indie devs, I’ve tried Copilot, Cursor, Claude… but when I wanted to build something real (like a web app with user auth or multi-module system), they just weren’t enough.
As indie hackers or small teams, we usually don’t have the budget for a big dev team — but we still want to ship products fast.
So I started building my own AI coding tool. My goals:
- Understand complex requirements written in natural language (not just single functions).
- Generate structured design + end-to-end runnable code, not just fragments.
- Help indie devs launch usable products faster, while keeping design and code aligned for easier maintenance.
Early result
I used it to build a booking system for a BnB. It was running in 2 days — something that would normally take me weeks. Still lots of debugging and architecture challenges ahead, but I’m super excited by the progress so far.
Next steps
- Improve accuracy from requirements → code.
- Add modular code generation for maintainability.
- Get more indie devs to try it and give real feedback.
👉 If you’ve ever tried Copilot/Claude/Cursor: what was the biggest pain point for you?
Would a tool like this actually help you ship faster — or am I just scratching my own itch?
Thanks for reading 🙏
I love using claude code, but there are some lack of features like in Warp, when i can see the code i have some decent code editor, all in one - maybe you are working on something similar? I would love to use some mix of Claude Code and Warp, if that's possible
Totally get what you mean! Thanks for the reply 😄. My tool focuses on generating end-to-end runnable code and handling software architecture. We’ve been grinding away recently, and I’m super excited for it to finally meet everyone!
Sounds inspiring. How the things are going so far?
Thanks for the reply 😄. We’ve been grinding away recently, and I’m super excited for it to finally meet everyone!
Really impressive work, Nio!
Building a tool that goes beyond snippets to generate end-to-end runnable code could be a real game-changer for indie devs. You might find Sky-T1-32B useful—it’s an open-source LLM with strong reasoning and coding capabilities that could help your tool handle complex requirements and modular code generation. I actually explored a similar topic on Medium here: https://medium.com/@sonuarticles74/why-saas-content-tools-still-cant-capture-your-voice-7a09cc5bfdff
There's something to building your own tools. How are you handling code editing? I actually built a toy version last weekend. My version is naive and reads/writes the entire files. This blows up the context window pretty quickly.
Yeah, even with today’s bigger context windows, it’s still not enough for serious coding. My approach is to break problems into small chunks so the AI only handles one at a time, and to use a graphical interface to make back-and-forth with the AI easier. I lean on software design to make this work — it’s easy to visualize for users and it naturally decouples requirements, which keeps each AI step small and focused. Lots of details behind this, but I’ll share more in a longer post later.
Really cool direction, Nio — love how you’re targeting the “ship something real” gap where Copilot/Claude/ Cursor still fall short.
Building a BnB booking system in 2 days is seriously impressive.
I’m curious: how are you handling debugging and long-term maintainability as you auto-generate more complex modules?
And do you see indie teams using this as their main dev flow or more as an acceleration tool alongside a traditional stack?
Excited to see how this evolves — please keep sharing updates 🙌
The core idea is to bring software design into the process. I’ve developed a framework for AI-assisted modeling and design. With the design in place, the engine guarantees that the framework code stays stable and consistent, so I only need to focus on reviewing and debugging the logic parts that are not tied to the design. This not only reduces the burden on the AI, but also makes it much easier for humans to read, debug, and maintain the system over the long term.
I tried base44 (sorry if you are looking for comments specifically about one of the 3 you listed above), and I find there is no modular approach in coding, which becomes inefficient in a way.
after all, can code fast, publish/ ship fast.
Back in my mind, another the potential pain is how to move the code out when i want to do so. I mean, using these platforms, this is the worry I have - the trouble of moving it out when you want to grow it big.
Building your own AI coding tool makes total sense, because it means you can tailor it exactly to the workflow you need rather than bending your process around generic AI. I’ve seen a similar approach in other niches too — for example, the PvZ Fusion APK community thrives because someone took the base game and built it out with hybrid plants and new mechanics the original devs never included. It’s the same principle: extending what exists so you can actually ship and scale.
hat’s a bold move, and honestly really inspiring 👏. I think a lot of indie devs can relate — tools like Copilot and Claude are great for snippets, but when it comes to building a full product with real-world complexity, they fall short.
Appreciate that 🙌. Yeah, I’ve felt the same pain — Copilot/Claude are awesome for little coding boosts, but once you’re trying to stitch together a real product, it’s a different game. That’s the gap I’m trying to close with this project — so indie devs can go from idea → working app without getting lost in boilerplate + messy architecture.
While it’s exciting to see how the Reverb Calculator has become a go-to tool for producers and engineers, the real value lies in the bigger picture. What challenges did creators face before such a tool existed, and how has it adapted to different music production needs over time? Understanding the journey behind its growth can offer useful insights for anyone aiming to create tighter, more professional mixes.
Great point — I see the parallel too. Reverb Calculator turned a messy, trial-and-error process into something clear and repeatable. That’s exactly what I’m aiming for in coding.
What AI tool was the best, or mayby use all of them
Exactly 😄. I have a friend who actually uses several of these tools at the same time. It gave me confidence for two reasons: first, developers are really willing to try new tools; second, it shows that if a tool solves a real pain point, there’s still plenty of room to make an impact.
Generating the initial scaffold isn't the hard part; it's living with it.
These tools are becoming powerful scaffolding cannons, erecting the skeleton of an app in hours. But the real work isn't framing the house; it's the plumbing, electrical, and fixing the inevitable leaks that appear months later.
The bottleneck simply shifts from creation to comprehension.
How do you ensure the generated architecture isn't a 'golden cage'—beautiful to start, but impossible to modify or escape when the product inevitably needs to pivot?
Totally agree — building the skeleton is just the start; the real challenge is keeping it maintainable over time.
Here’s what we’re trying with our AI tool:
·Treat modeling as coding — AI makes it easy, and the model reflects the code accurately
·Visual links to help devs quickly understand the overall design and pinpoint issues
·Clear domain/module separation, independent read/write views, and modular code generation so each component can be understood and updated independently
Curious — have you tried or seen any approaches to avoid generated code becoming a “golden cage”?
Great question. The most effective approach I've seen is to stop treating the AI as a code generator and start treating it as an infinitely fast, slightly naive junior developer.
It needs a 'senior dev' (us) to provide a strict 'style guide'—not for syntax, but for architecture. We define the non-negotiable patterns, the module boundaries, and the 'don't-cross-this-line' rules before it writes a single line.
The goal then isn't a perfect first draft, but a first draft that's already sitting within a sane, human-approved architecture. It's about containment, not just creation.