Software Development and AI

This blog was contributed by Marc-Alexandre Cartiant, Solution Architect Lead - Data Science at ADI.

I've spent more time coding in the past two years than I had in my entire life before.

This is because ChatGPT has enabled something remarkable: converting natural language specifications into working code. While it’s not always perfect and can be buggy, a software engineering background helps spot and fix issues quickly.

With solid experience, you know what to ask, how to ask, and how to manage various parts of a program, enabling faster builds.

So why do we still need developers if we can get results so quickly and easily?

The market standard for digital products will rise. Individuals capable of replacing entire teams will become more common, but that also means projects must be more ambitious, pushing boundaries to stand out. This demand is driven by market dynamics.

Ironically, I remember being 13 or 14, reading a computer magazine predicting the disappearance of software developers—a recurring narrative that resurfaces every few years. Shortly after, however, there’s always talk of a "shortage" of engineers.

Experimenting with AI to build prototypes has become straightforward, especially if you have a stack that allows you to perform multiple roles. AI enables one person to handle simple use cases, akin to the early internet era when building a simple webpage gave you status. It required some technical knowledge—just enough to make an impression.

I think we’re ending this initial phase of LLMs and moving toward more sophisticated implementations, like when static websites evolved to dynamic ones requiring databases. In the LLM context, the equivalent of a “database” is RAG (Retrieval-Augmented Generation). RAG behaves like a super search engine across large document sets (thousands or more), cross-referencing documents to extract relevant data based on queries.

Put differently, RAG becomes an advanced AI component, enabling new use cases such as generating documents from templates that reuse stored information.

This transition resembles the shift from "building a static webpage" to "building a dynamic website with a database." Implementing RAGs requires more technical expertise, marking it as phase two.

Phase three will see even more ambitious projects, demanding stronger tech stacks to deliver innovative services. Websites like Google and social media are examples of highly complex technical architectures.

To revisit my initial point, I believe we’ll see a significant leap in delivering digital solutions; more smart features built faster will become common. Yet, I doubt we’ll see the sudden end of engineers or coding. Historically, such predictions have been incorrect.

(This blog was originally published by Marc on Work with GenAI.)

About Marc: Marc has over 20 years of experience in digital transformation and a specialization in Large Language Models (LLMs) since 2022. He's helping companies to embrace GenAI, starting small and simple and then integrating one by one critical business processes.