Opinions on AI as a tool

Wondering how you can support this newsletter? I would love it if you bought me a coffee. It helps me write more code and more newsletters.

I’ve spent some time away from writing on this newsletter for personal reasons.

But,

we are so back.

I have a plethora of ideas to write about and I’m excited to bring them to you each week.

Hold on to your pants, let’s talk about AI.

Almost no facts, only a few opinions to bless your brain cells.

Past

It has been roughly 2 years since ChatGPT was released to the public (feeling old yet?). When it was first released, I recall everyone going buck-wild about it and thinking that it was going to take over the software engineering world. Everyone would lose their jobs and we’d all be slaves to the almighty LLM.

I’m not going to act like I wasn’t a doomsayer, because I sort of was. I didn’t necessarily think that we would all be cooked, but I did think that many people would lose their jobs because of it. I can’t say I told you so, because I wasn’t writing then, but I was pretty correct in that regard. Many junior engineers did lose their job during the past few years. It’s okay though, most of them found now jobs a few months later when the hype died down. If you were affected by that, I feel for you and hope you’re back in a role. If you’re not, my bad fam.

It wasn’t just ChatGPT that did this. The economic situation in the world at the time wasn’t necessarily stable. Interest rates were rising after the short-lived boom after COVID and investors were pulling back from startups. I know this first-hand because a client of mine laid off over 30% of the company due to funds not wanting to continue investing.

Enough of this talk though, let’s focus on the tool.

In the beginning, with GPT-3, it was kind of like your friend that can solve a rubik’s cube in 2 minutes. It’s pretty cool and reasonably fast, but who cares? It worked about 40% of the time to get you a reasonable solution, but 40% isn’t really that great. So why use it? For me, I tried it for about 2 months, improved on my prompt writing, still couldn’t get the right results, and then stopped using it for a while.

A little bit later, Copilot shows up like this:

Copilot was also pretty cool at the time, but honestly, hitting about the same rate as ChatGPT. The cool thing was auto-completion. However, sometimes it felt a bit too aggressive in how it made decisions. Almost like it was trying to dictate the code I was writing and not me. So I adjusted some settings and ultimately it was just a nifty autocomplete tool that still caused more trouble than good. This ended up with me cancelling copilot about 2 weeks after starting it as it was actually inhibiting my workflow.

To be real with you, ChatGPT and Copilot kind of sucked 6 months after release. Fast forward to early 2024 though and Sam Altman & OpenAI release GPT-4o. 4o for me was when it started to become an integral part of my work. I was getting more and more accurate answers with ChatGPT on complex topics & Copilot starts to become more and more accurate with it’s autocomplete suggestions.

Present

Fast forward to the present day and I am invested. I use copilot on the regular and use ChatGPT often for complex chat conversations. For instance, I use ChatGPT to answer technically complex architectural questions, write complex SQL queries, and also to give me recipes based on what I have in my kitchen at the moment. This is peak technological innovations.

I use copilot to autocomplete what I’m already thinking in my head because 80% of the time I’m just pressing tab. I have noticed that I find myself writing my function names as if they are a prompt as a function name. Doing so gives copilot some context as to what the function will do and since it has already injected every company’s intellectual property into its mechanical veins, it already knows the O(1) or O(log n) solution for that and I just press tab. Tab this, tab that. That’s how the average software developer’s days are now. It’s almost a meme.

It’s brilliant, but it does make me question. Is it easy only because I’ve been in the industry for some time and I know how to easily determine if the autocomplete result is the proper solution? Or is it really that easy to be a developer now? I actually don’t know, but would love to hear from anyone that is starting out how they feel. Feel free to write to me what you think.

Future

I’ll give a bold statement here and I’ll say this:

Based on my knowledge of LLM systems and how they are used today, my experience in the industry, and my personal beliefs.

I don’t think software engineers will be out of a job any time soon.

We build software to fix people’s problems. There will always be a need for someone to fix someone else’s problem with technology. We’ve been doing this for thousands of years not just on computers.

I do think that the average engineer will need to know much less about the deep, technical knowledge of how computers work though. However, this is also produces areas of opportunity for engineers to carve a niche for themselves. The average engineer will do their comfortable work at a cushy enterprise business, but the scrappy, in-depth knowledgable ones will still be building deeply technical solutions in startups. The common denominator is that the tools will be the same, but the prompts will be different.

Away from the people aspect, technically, I think that we will continue to see trends to individualized models for different industries. I’ve seen many startups in various industries that have niched down their models to work with a single solution. While ChatGPT is a great tool, it’s limited in that it has general knowledge. Niche solutions are able to provide more valuable insights for specific industries. I see tons of opportunity here since there are essentially infinite industries and niches in the world today.

A bit of a rant on this one, but after some time away, it feels good to write about my experience with AI and the tooling around it. I’m by no means a deep technical expert in this field. However, after working in the field with these tools, I’m quite comfortable with them and it is now a part of my daily routine.

If you enjoyed this content, please share it with a friend, or buy me a coffee with the link at the start of the article.

Thanks for reading!