Learnings From Our AI Assisted Software Development Summit
AI assisted software development is undeniably already changing the way teams work. At its simplest it is autocomplete on steroids. On the other extreme, it is full-on vibe coding.
We held an AI assisted software development summit at our office last week. Because we like to be bold, we titled this event Tech & Consulting: Vibe Coding, a name sure to ruffle some purist feathers.
Alongside our own team, we were joined by a panel of external experts: Petri Louhelainen from AgileDay, freelance consultant Juha Litola and Jesse Luoto from Reaktor.
-
Category:
Developing -
Author:
Witted Megacorp -
Published:
Great Leaps for Individuals, Small Steps for Teams
We’re taking great leaps in productivity individually thanks to AI assisted software development. Anyone using LLMs will be churning out more code, writing tests faster than before, and being more productive on individual tasks. Still, on the team level, we’re lagging behind as the product level best practices aren’t there yet.
Less Coding, More Guiding & Validating
AI has taken some coding away from coding. Software professionals architect, guide the development, test and validate outcomes more, and thus, they code less. Still, the fundamentals of great product development haven’t changed: you need great people, cross-functional teams, and good communication.
Mo’ Power, Mo’ Problems
There are new topics to address that arise from the advent of AI assisted coding.
Security practices need to match also new threats such as prompt injection and other vulnerabilities.
The tech stack selection to be based on average quality of the available codebase used in training the models. Not pointing fingers, but sorry, no PHP!
The documentation of the logic and role of components becomes even more important as AI can create code with a very complex logic unless properly guided.
With great power comes great responsibility
AI needs guardrails. If it starts losing its grip, it’ll go off the cliff quickly. And that’s a steep cliff. You need to have a solid architecture, good documentation, clear implementation instructions and validation.
Is it getting better?
Even at the risk of sounding too optimistic, we have to say it’s all getting better. The ever-bigger context windows help finding needles in even bigger haystacks. The technologies are improving fast, and so will productivity.
To sum up, the software development landscape is changing rapidly. Some basic tenets still persist: Developing software in a team was very much about communication earlier, and it will be even more about it in the future.
