It s the third week of working on my little side project, SimploMail.com, and to speed things up I ve been doing a lot of vibe coding. It s been fun, but a few things became obvious pretty quickly.
I stopped using the auto model setting in the IDE. When it silently switches models, the quality drops fast. I can feel when the agent all of a sudden looses its intelligence . So now I just pick one model I trust and stick with it.
I also try to keep each AI session focused on one small task. One feature, one change. After it writes the code, I go through everything myself. I check for hard coded config, make sure it didn't quietly delete a unit test to make something compile, and etc. Sometimes it does update unit test just to make it pass .
And I never commit without reviewing. The AI is helpful, but it can also remove something important without telling you. I've seen it happen enough times now .

After rebuilding the same project three times because AI forgot my architecture, I got fed up and built @CodeRide (Beta) with my team.

The problem: AI code assistants lose track of your project between sessions. Every time I start coding with Cursor, Claude, or any AI assistant, I waste time re-explaining my codebase structure, architectural decisions, and coding patterns.

What we built: The project management tool for coding agents using MCP. Upload your project documentation or PRD, and CodeRide breaks it into optimized, fully contextual tasks ready for your AI agent.

I've been talking to dozens of PMs over the last few weeks who prototype with Lovable, Bolt, Figma Make, V0, etc. Same frustrations keep surfacing.

• Output looks a bit generic: looks like a demo, not your actual product

• Context loss: explain your product in ChatGPT/Claude, then re-explain in Lovable, then again somewhere else

• No edge case thinking: AI executes prompts literally, doesn't challenge or expand on them

The core issue I keep seeing: these tools are interface builders. They're great when you already know exactly what to build. But the hard part, thinking through the flows, the states, the edge cases, where users will actually get stuck, that's still entirely on you.

I ve been experimenting with a CLI workflow that removes the most annoying part of AI-assisted dev:
re-explaining context every single time.

Instead of prompts like:

Integrate Stripe payment gateway with X, Y, Z

I just run:

I'm Kevin, founder of EnginifyAI. Over the last 12 months I've been working on a prompt engineering tool. I started off using Bolt but quickly I found myself unable to completely use it effectively. I then tried Cursor and was using it early on but maybe I had the wrong settings as I found myself using all my credits in 2 days. So, I switched to Windsurf, and to my surprise credits were a bit more manageable and it took about 2 weeks before my credits were consumed. Keep in mind that I am new to coding so I spent a lot of time asking questions in Windsurf instead of using my paid subscription to Anthropic.
I'm coding using React/Typescript and Supabase. I have learned that coding is a lot of work, a lot of fine-tuning, and a lot of testing and fixing. The one thing I haven't quite figured out completely is how to fix migrations. Early on I was running SQL scripts directly in Supabase without migration scripts because I found it to be faster. But now Windsurf uses very old migration script references and I find myself fixing the same issues because of an old migration script reference. Any idea how to fix this?
I assume that vibecoding is an experience everyone can enjoy but comes with frustrations about asking the right questions, or being very specific about my requests. I am using Claude Sonnet 4, 4.5 and 4.5 thinking. Are these the best coding ai models for Typescript?

Do any of the more experienced coders ever feel like closing the loop in vibecoding is a bit too much control to let go off? I'm not sure if I can categorize myself as an experienced coder. I'm been mainly self-taught and been doing it for almost a decade now, so I have strong opinions about what I consider best practices. (I actually love refactoring, because it feels like order is once again restored in the codebase).

However, with AI becoming so good at so many parts of the product development, I'm starting to feel like a project manager, or god forbid, scrum master... On the one hand, it's nice that I get to spend most of my time thinking about the problems themselves rather than the implementation of the solution. But on the other hand, I think my coding style is a bit like trying to find a way out of a dark room by running into the wall repeatedly as opposed to planning a way out. And I kind of missing running into the wall to be honest. Now, I'm enviously looking at AI hitting the wall.

I'm what you might call a gringo vibe coder.

No dev degree. No traditional background. Just me, Cursor, Claude, and AI... figuring things out together.

Rick Rubin said it best: "In the past, for music, you had to go to the conservatory and study for years... then when punk rock came along... if you had something to say, you could say it." He calls vibe coding the punk rock of coding.

I've been vibe coding for a few months as a non-coder and I'm still annoyed with the fact that I can't understand what's happening under the hood.

I've got a decent understanding of code but I can't actually write it and I don't know the best practices for stuff like architecture and security which apparently aren't baked into most vibe coding tools. So my question is...

I am seeing this with Lovable, Verdent, Replit: I am clear, RAG my prompts, clean them with Claude Opus 4.6, find that with tough problems, they'll tell me something is fixed or done, and it isn't. Not even close. And I'm burning costly credits, making no progress.
I'd love to learn what others are doing.
I'm building a multi-agent tool that integrates with 25+ LCNC sites and also IDEs, and yes, there are tough problems with the tool's awareness of what a user is doing in a console or fields in a browser window. I'd apprecaite it if the nocode platform simplay told me, "no can do," rather than trying one thing after another, staying stuck, and costling me $100 or more per day to keep failing.
Thanks, in advance, for your suggestions!