Using ChatGPT: Like An Upgrade From a Normal Bike to a Powerful E-Tandem
Working with ChatGPT: It's A Productivity Upgrade 🍄🎮🌟:
So, I've been using ChatGPT to help me experiment and create with ThreeJS, and overall, it's like upgrading from a classic Dutch bike that you're riding up a hill in the rain:
to a powerful e-bike ⚡ with an over-eager junior engineer 👩🔬 up front that you're riding through beautiful French countryside:
A Definite Increase in Productivity: 📈
I could have implemented the ThreeJS things that I have so far with without ChatGPT, but it would have taken me a month or so, involved hours of StackOverflow, Reddit, Googling, and generally been slow and quite painful 😫. So, using ChatGPT most definitely has resulted in a significant increase in my productivity. However, there are a couple of caveats ....
You Still Need to Know What You're Doing:
ChatGPT frequently makes suggestions that you just have to check, and to do the checking you have to know all the usual stuff about coding: what things to look out for (e.g. won't scale, poorly written, inefficient, insecure), how to debug the code yourself, and be able to tell what things ChatGPT has just missed or got wrong,
Simply copying and pasting the code I'd say is a really bad idea. 👎
Discipline is Required To Resist ChatGPT's Constant Suggestions:
The other thing that's been really surprising is that ChatGPT is like a really eager not-so-experienced engineer:
- It makes lots of suggestions.
- Some of those suggestions just don't work, and despite interacting repeatedly with ChatGPT, after a while you realise just won't ever work, and ChatGPT won't admit it. I disappeared down at least two rabbit holes 🐰 🕳️lasting over an hour where it never said "Ah, you got me, I don't know how to do this", but it became pretty clear that it just didn't have a clue. So, I had to back-out and try something else.
- It makes suggestions beyond your initial goal, so unless you're really careful you spend time over-delivering. It's very very tempting to find yourself thinking "oh, that's a nice suggestion, if you're offering such a solution, don't mind if I do" and before you know it, you've put a significant amount of time into something unnecessary. I can imagine in a tight-schedule environment (are there any that aren't?) trying to be purely Agile and delivering that tiny slice of value to get feedback may require some discipline to stick to.
Need Less Engineers - or Upgrade Your Engineers?
In short: a clear increase in productivity but with some things to watch out for. So, does this mean that an organisation can do with less engineers? At the moment, right now, I would argue not : essentially, organisations are in a race with no rules - if your competitors have upgraded to GenAI tooling and you haven't, you are more likely to lose - they'll deliver capabilities sooner and you risk losing customers/clients.
Obviously it depends what you're trying to achieve: if your organisation is prioritising finance over delivery, then that's a different calculation.
Few Final Notes:
- This is a solid in-depth analysis of working with AI tools in engineering: Pragmatic Engineer: Learnings from two years of using AI tools for software engineering - says similar things but in greater depth.
- This world is moving so fast that there are now GenAI tools that totally allow you to "vibe code" i.e. create something purely from natural language e.g. see The Blueprint AI: Anthropic now lets you make apps right from its Claude AI chatbot and Lovable.dev. For now, for my purposes I'll stick to working away with ChatGPT but these things are definitely worth keeping an eye on.