If you’ve opened twitter at least once in the last 2 months you would have already heard about vibe coding.
Vibe coding is where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. LLMs (especially Cursor) make this possible.
I "Accept All" always, I don't read the diffs anymore. When I get error messages I just copy paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension, I'd have to really read through it for a while.
~ Andrej Karpathy (arguably the most famous Slovak person after me).
It must be worth our attention if one of the key people in developing the entirety of AI plays around with this approach.
It comes with many pros/cons that need addressing.
It excels when creating something from scratch.
From scratch + having a rough vision of a similar product in mind speeds up the process exponentially. Feed him a product you want to re-create step by step, you will get a competitor’s product/feature done very quickly. After that you can address the limitations it might have and perfect it.Not everyone is familiar with code environment.
Coding is getting popular by the year, however most people are not familiar with debugging/correctly formatted prompts. This gives experienced developers competitive edge and saves time on finishing vibe-coded products.Shifts the entirety of software engineering.
Most of the vibe code works, even though it’s spaghetti code. The larger the codebase gets, the more difficult it will be to maintain it. This splits future developers into two categories. A) cheeky visionaries that get the ball rolling and B) the maintainers that come in later and try to fix and maintain the codebase. Currently the average software engineering team consists of 6 engineers, 1 PM, 1 BA, and some extras. This might become 2:1:1:3(designers).Coding part is not the main problem. It is means to an end.
Many people think coding is a god-like skill when you achieve it, it sets you for life. In my opinion, coding is secondary. First is trying to understand and break down complex problems into smallest possible chunks (obviously). Solving a problem requires a certain out-of-box-thinking with understanding where the edge cases may lie. Cursor is not going to cover those edge-cases until you necessarily ask it to.Smaller, yet more efficient teams.
In past you needed a team of ~10+ people to achieve figures ~1M ARR and it was already tough. There’s more and more startups emerging with less employees and higher revenues.
Here’s a step-by-step tutorial even your grandma could use to create her first product:
Generate (detailed) project specification using Grok
Ask Grok for the best tech stack for the given project specification + add the cursor-rules (such as “keep things simple”)
Chunk down project specification into small tasks with each consisting of super easy steps to complete
Add visuals you liked from Pinterest, competing products to your Cursor prompts.
Cursor with Claude 3.7/3.5 when you run out to build task step by step.
New chat after a task was done to limit confusion.
When things don’t work, prompt away to debug, if used obscure, outdated libraries - > ask to approach differently
Tadaa, enjoy your new friend Cursor for the next 1,000+ chat iterations.
Please mind the bigger the code gets, the more complex debugging. Good approach would be:
Build a great MVP for a problem that needs solving
Raise a round/angel/get into accelerator
Hire a senior engineer that fixes your code
How I can help
Software
I co-own a software house with 155+ software developers. Need a product built? Message me and I’ll get you a discount!
Advertising
Advertise in my newsletter to get in front of 10,000+ founders.