02/03/2026 1:30 am
Topic starter
Notifications
Clear all
Other LLMs Forum
6
Posts
7
Users
0
Reactions
254
Views
0
honestly im just about at my wits end here because ive been trying to get this delivery app project off the ground for my small business in Seattle and everything i touch feels like its breaking. i have like two weeks to get a working MVP for these beta testers and my current workflow is just... its a disaster. ive been using github copilot because everyone said it was the gold standard but lately it just feels like its hallucinating half the functions and im spending more time debugging the automated code than i would have spent just writing it myself from scratch. its so frustrating because im not a senior dev by any means so im relying on these tools to bridge the gap but theyre just leading me into circles.
my logic was that if i paid for the pro version it would actually understand my local files better but it keeps suggesting libraries that dont even exist or deprecated methods from five years ago. i keep hearing people talk about cursor or maybe using something more like an agent like devin or opendevin but im nervous about the cost because my budget is strictly under $100 a month for tools right now. i was thinking maybe claude 3.5 sonnet inside an ide is better? or should i be looking at those specialized autonomous agents that claim they can fix bugs while you sleep? i really need something that can actually handle full context across multiple files because jumping between tabs and pasting code is driving me insane.
i just need to know what people are actually using for real production work when theyre in a time crunch. is there something that actually understands the architecture of a whole project instead of just guessing the next line? im literally losing sleep over this deadline and i feel like im fighting the ai instead of it helping me. if anyone has actually switched away from the basic stuff to something more pro that actually works for building full apps let me know. i cant keep wasting hours on these weird logic loops the current setup keeps giving me...
6 Answers
12
02/03/2026 3:15 am
To add to the point above: I actually disagree with switching editors again. Most plugins struggle with deep context across multiple files. Try Aider AI CLI Coding Assistant using the OpenAI GPT-4o API. It indexes your project properly so it doesnt hallucinate nonexistent libraries or old methods as often.
- Maps your entire codebase structure.
- Handles multi-file refactoring via terminal. Its slightly more technical but much more reliable for hitting a deadline.
10
02/03/2026 1:45 am
Re: "I went through the exact same thing a..." - are you building web or mobile? I'm happy with these:
- Sourcegraph Cody Pro for context
- Supermaven Pro for speed
2
02/03/2026 1:31 am
I went through the exact same thing a few months back while trying to launch a logistics tracker on a tight budget. Copilot was basically just a fancy autocomplete that kept tripping over itself. I switched to Cursor AI Code Editor Pro and it was a night and day difference because it actually indexes your entire folder. It uses RAG to understand how your components interact, which sounds like exactly what you need for that Seattle delivery app. Heres what I recommend based on my own experience:
- Get Cursor AI Code Editor Pro for $20 a month.
- Set the model to Anthropic Claude 3.5 Sonnet inside the settings.
- Use the @Codebase feature to ask questions about your whole project. The chat feature in Cursor actually reads your files and wont suggest those hallucinated libraries as much. Honestly, agents like Devin are still too pricey and unpredictable for a two-week deadline. Stick to a solid IDE that knows your context and youll probably hit that MVP target without losing more sleep.
2
02/03/2026 2:31 am
🙌
1
02/03/2026 1:55 am
Noted!
1
02/03/2026 3:40 am
I went through a similar nightmare last year while trying to launch a client site. Spent way too much on subscriptions only to have the AI hallucinate half the backend logic. One thing I would suggest is to be really careful with those autonomous agents... they can chew through your budget faster than you think if they get stuck in a logic loop. I once burned $40 in like an hour because a tool kept retrying a broken test. Honestly I found that using Aider AI CLI Tool along with the DeepSeek Coder V2 API was the best way to keep costs down. Since you are on a $100 budget, paying for tokens via something like OpenRouter is way smarter than a flat monthly fee. It handles the whole file tree pretty well too without the constant copy-pasting. Just watch your API limits tho so you dont get a surprise bill.
