From idea to MVP with GenAI: this is how we do it at Blenddata

The days when developers routinely go to Stack Overflow when they havea bug or problem is over. Fewer and fewer people still write code without the use of GenAI. In just a few years, the use of GenAI tools has gone from unknown to indispensable, even in our way of working. It is almost impossible to imagine that when Blenddata was founded, these tools were not yet publicly available. Realise that we are then talking about 2022.

By now, ChatGPT, Cursor and other tools have been woven into the way Blenddata’s developers work. In this blog, we take stock: how do you properly deploy GenAI to help you write code without compromising on quality, portability and craftsmanship?

What is GenAI and how exactly does it work?

Generative Artificial Intelligence (GenAI) models are capable of generating an answer based on a textual question. These types of ‘chatbots’ are trained on billions of pieces of text to learn the underlying patterns and relationships between words and phrases. These patterns can be used by GenAI, based on input (a question), to produce new text (an answer).

As a chatbot learns the connection that water is wet in this rich set of data over and over again, it will respond to the question “Is water wet?” with “yes”. Not because it conceptually understands that water at room temperature is a liquid, but because it has encountered the connection between the words “water” and “wet” countless times in the text it has been trained on.

The text on which these chatbots have been trained comes mainly from the internet, which may also include the 24 million questions and 36 million answers found on Stack Overflow, or code snippets once posted on Reddit. This also enables GenAI to generate or review code: those patterns are also known to the chatbot.

How does Blenddata deploy GenAI when developing code?

• Writing code: A simple SQL query? Python code needed to read an API? A PowerShell script to rename all the files in a folder? If you need fast working code or want to validate a concept, ask GenAI. At a hackathon, writing code to test your idea is no longer the limiting factor. By making smart use of GenAI, going from an idea to an MVP has never been easier.
• Debugging: your code returns an error. Do you then go looking in the package’s documentation? Or to Stack Overflow to see if someone else already had this problem? Probably not. By providing GenAI with the full stack trace, you almost always find out the cause of your bug within seconds.
• Code review and bug detection: We apply the four-eye principle: code should always be reviewed by at least one other developer before it goes to production. GenAI does not replace this principle, but it can add an extra layer of review. When you ask GenAI to review your code, it often sees things you haven’t considered, no matter how experienced you are.
• Refactor and improve existing code: Over time, any codebase can get messy. GenAI can help restructure and rewrite code to make it understandable and well-constructed again. If you have written good tests, you will immediately notice if the code is still working correctly.
• Help in understanding code: Do you come across a function of 100 lines without a docstring called process_data? Then ask a GenAI to explain or comment on the code. This way, you can more quickly assess whether the function actually does what it is supposed to do within the business context.

The advantages and disadvantages of using GenAI