Vibe Coding continues to gain ground in Silicon Valley, and the old CEO of Monzo, Tom Blomfield, has reflections on how to maximize his potential.
Invented only two months ago by Andrej Karpathy, an OpenAi co -founder, the term refers to people using AI to write code by giving it textual instructions.
Experienced engineers use it to save time, and those who have non-technical backgrounds code everything, game dating applications.
Blomfield, who is now a group partner at Y Combinator, shared some tips for people looking to optimize how they vibrate the code, in a video published by the accelerator on Friday. Here are three tips he gave.
Choose the right tool and create a full plan
BLOMFIELD advised users to plan in advance and to experiment to find the tool that best takes their skill level and the desired final product.
He noted that tools like Lovable and Replais were adapted to beginners, while the more experienced coders could use the board in the board or cursor.
“Work with the LLM to create a full plan,” he said in the video, referring to major language models. “Put this in a Markdown file in your project file and continue to refer to it.”
He suggested that users could use the LLM to perform the Plan section by section, instead of making the product at once.
“This advice could change in one or two months, because the models improve,” he added.
Do version tests on the product
BLOMFIELD said that when he had encouraged the tools of AI several times to the same coding task, he would get bad results due to the model accumulating “bad code layers”.
Related stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
He advised to use the large language model to write tests that simulate someone by clicking on a version of the site or application, to assess the way the functionalities work.
Sometimes LLM can make unnecessary changes to these features, he said, and the implementation of integration tests can resume these changes more quickly.
Write instructions for LLM
Blomfield said he had found that different models had succeeded where others had failed. If a user meets a specific bug, it is useful to reset all the changes and give the detailed LLM instructions to repair it on a clean code base.
“Journalization is your friend,” said Blomfield.
Another advice he offered was to use small files and a more modular architecture based on service, where the LLM has clear API limits.
An advantage of this is that it would avoid creating a huge single standard of code for various projects, which could be more complex to manage and have more integration challenges.
businessinsider
