+- Hack Forums (https://hackforums.net)
+-- Forum: Tech (https://hackforums.net/forumdisplay.php?fid=444)
+--- Forum: Artificial Intelligence (https://hackforums.net/forumdisplay.php?fid=460)
+---- Forum: AI Programming and Vibe Coding (https://hackforums.net/forumdisplay.php?fid=462)
+---- Thread: Tips on building an SaaS with Claude/Cursor? (/showthread.php?tid=6325696)
Tips on building an SaaS with Claude/Cursor? - l7irewolf - May 19, 2026
I have everything planned out on how I would like the whole software to be, UI/UX designs on Figma ready...
I have no actual coding experience, so I thought of using Claude Code + Cursor to build it for me.
How good is it at building if my prompts are non-technical, but descriptive of what each function should be?
Any tips or resources besides Youtube I should look into?
Anyone else built an CRM/ERP with such tools?
RE: Tips on building an SaaS with Claude/Cursor? - Omniscient - May 19, 2026
Have you considered OpenAI's Codex yet? Currently being heralded as the best.
RE: Tips on building an SaaS with Claude/Cursor? - l7irewolf - May 19, 2026
(May 19, 2026 11:35 AM)Omniscient Wrote: Have you considered OpenAI's Codex yet? Currently being heralded as the best.
I haven't. I will look into this option. I appreciate your help!
RE: Tips on building an SaaS with Claude/Cursor? - eXceptional - May 19, 2026
If you're using Claude code using CLI, consider installing superpower plugin and some others, like code review.
You will save on tokens and there are some agentic plugins too.
I prefer Claude Opus 4.7 over Codex.
RE: Tips on building an SaaS with Claude/Cursor? - 5chn4pp - May 19, 2026
Personally I use Codex with GPT-5.5 extra high. Also the projects feature in Codex helps a lot.
I ditched Claude because it's limits are just too high. I can use all my tokens in 30 minutes with Claude. With Codex I can marathon code for 12 hours and not get limited.
RE: Tips on building an SaaS with Claude/Cursor? - Binance - May 20, 2026
(May 19, 2026 01:04 PM)5chn4pp Wrote: I ditched Claude because it's limits are just too high. I can use all my tokens in 30 minutes with Claude.
It's literally so bad, im shocked people can use this tool. It continuously drops task and ruined momentum.
RE: Tips on building an SaaS with Claude/Cursor? - Prawnster - May 20, 2026
The few times I've used Claude, usage seems to be gone in about 15-20 mins (only on the pro plan).
I'd recommend Codex if you want to vibecode.
RE: Tips on building an SaaS with Claude/Cursor? - JDAM - May 20, 2026
(May 19, 2026 11:35 AM)Omniscient Wrote: Have you considered OpenAI's Codex yet? Currently being heralded as the best.
My understanding is that Claude is the most popular because it's the best at compensating for a lack of detail and deliberate instruction.
In other words, it's better at "vibecoding" for people who don't understand programming. For people like me, it just werks... more often than ChatGPT
ChatGPT is, across the board, not as effective at making up for inadequate instructions. Opus 4.7 is also worse at this than Opus 4.6, hence why Claude's subreddit is full of people posting about how 4.7 is actually worse than 4.6 for reasons other than its increased token usage. All other criticisms of Anthropic, such as the attempt to force people to use the Agent SDK with API pricing rather than their subscription-based usage (even with the free credit), are entirely valid.
OpenAI hasn't even released a 5.4 or 5.5 Codex model. Maybe because 5.3-Codex is good enough, 5.4-mini is a good-enough substitute, or they want to get everyone on 5.5, which has higher API pricing than Opus 4.7.
To answer OP's question, he probably won't end up with a product worth purchasing, regardless of what he uses, but there isn't much point in using Cursor if he lacks programming knowledge and isn't planning to learn. Cursor offers less usage per dollar, but offers a wider range of models and their own a VSCode fork built around their service. I have a friend who uses GPT-5.3-Codex through Cursor for work and does very well with it, but he's an actual programmer and gives very deliberate instructions and explanations that require the knowledge he has to actually get the LLM to work the way he wants. ChatGPT's subscription usage limits are more generous than Claude's. There is a 50% extra weekly usage "event" for Claude through July 13th, but YMMV
Also, yeah, "Codex" is OpenAI's desktop app, comparable to Claude Desktop and also a set of coding-specific models, separate from their frontier models:
https://developers.openai.com/api/docs/models/all
RE: Tips on building an SaaS with Claude/Cursor? - Omniscient - May 20, 2026
(May 20, 2026 12:55 PM)JDAM Wrote: My understanding is that Claude is the most popular because it's the best at compensating for a lack of detail and deliberate instruction.
In other words, it's better at "vibecoding" for people who don't understand programming. For people like me, it just werks... more often than ChatGPT
ChatGPT is, across the board, not as effective at making up for inadequate instructions. Opus 4.7 is also worse at this than Opus 4.6, hence why Claude's subreddit is full of people posting about how 4.7 is actually worse than 4.6 for reasons other than its increased token usage. All other criticisms of Anthropic, such as the attempt to force people to use the Agent SDK with API pricing rather than their subscription-based usage (even with the free credit), are entirely valid.
OpenAI hasn't even released a 5.4 or 5.5 Codex model. Maybe because 5.3-Codex is good enough, 5.4-mini is a good-enough substitute, or they want to get everyone on 5.5, which has higher API pricing than Opus 4.7.
To answer OP's question, he probably won't end up with a product worth purchasing, regardless of what he uses, but there isn't much point in using Cursor if he lacks programming knowledge and isn't planning to learn. Cursor offers less usage per dollar, but offers a wider range of models and their own a VSCode fork built around their service. I have a friend who uses GPT-5.3-Codex through Cursor for work and does very well with it, but he's an actual programmer and gives very deliberate instructions and explanations that require the knowledge he has to actually get the LLM to work the way he wants.
There is a constant one-up between the AI's. It's going to be like that for a while.
RE: Tips on building an SaaS with Claude/Cursor? - JDAM - May 20, 2026
(May 20, 2026 12:56 PM)Omniscient Wrote: There is a constant one-up between the AI's. It's going to be like that for a while.
There are many different levers these companies can pull, and unfortunately, many of them are just "use more resources for better results."
I think we're out of good data, and now we have models built on increasingly wasteful amounts of self-sucking and compute hoarding.