LLMs Have Legitimately Changed My Life

LLMs Have Legitimately Changed My Life

I should say, specifically Gemini and Claude Code have legitimately changed my life.

When the whole chatGPT explosion happened, I remember being a bit confused why people were so excited about it. Google had been working on amazing AI tech for years. Some of which they abandoned, and are now kinda bringing back. Things like google now on tap. At the time, the AI “thing” was machine learning, which like, aren’t LLMs basically just machine learning?

Anyway, so I wasn’t impressed with ChatGPT especially because of how wrong it was about SO MANY things initially. Then google released Bard (later to be renamed to Gemini for some unexplained reason). It also made a lot of mistakes. I started to kinda doubt the technology. It didn’t seem worth the potential and very real impact on the environment, and the MASSIVE AMMOUNT THEFT!

I still don’t think it’ll ever be worth that, but these tools have become legitimately useful in my everyday life.

At some point, it crossed the line and became useful, but still felt distant from me.

And then I got access Claude Code at work. When I learned this app was a CLI tool, I was intrigued because there’s nothing that says “This is for software engineers” like building something for the Terminal. I think the real difference that Anthropic made was that they actually built software to address a core audience.

It was impressive, but not mind-blowing. Not until I watched a YouTube video and Anthropic introduced the skills feature, which was like creating little minion pieces of instruction sets for specified tasks.

The first thing I used it to do was automate our dependabot merges. If you know what those are, you know they are SUCH A CHORE. Claude has basically taken over that task for me, and it’s such low risk of a task.

Today, Claude Code is a full coding assistant for me. With enough examples and clear instructions, it can save you a massive amount of time. Time trying to find the right documentation. Time tweaking configurations in 18 different ways because the documentation is always too vague! It saves you time creating huge test objects with realistic test values. Time reading long complex stack traces or formating data!

It also makes me very sad.

On one hand, for businesses, this is an obvious win. If you are a small shop, your engineers just got at least a little faster, and more likely a lot faster, and you just had to add an extra $200 ish /month license for each of them. From a company perspective it is a huge cost benefits.

But for engineers, I would argue it’s probably not great. With great power, comes great responsibility. There are two versions of this.

Version 1: With great power, comes great responsibility:

First, the great power is the efficiency and speed you’ve gained, and the responsibility is the amount of work your company is now going to expect you to complete in a sprint. They will expect either a higher quality of work, a faster delivery, or both!. I’ve already seen this start to happen within the software engineering community. Historically, companies have not compensated workers more when they become more efficient. It would be nice, if the companies said:

“hey, because you have this tool that makes you so much more efficient, we will now have 4-day weekends!”

If anything, the opposite is happening, we’re losing our jobs!

Version 2: With great power, comes great responsibility.

Second, the great power is the ability to just have code created for you. Why even plan or design features, just have Claude figure it out. Grab your jira ticket, throw it into Claude, and chime in every once in a while to check on it. Raise the PR at the end and someone else can catch its bugs. The responsibility is to NOT do this!

In the content space, the creators that spent time to build trust with their audience, and put out high quality/trustworthy content don’t seem impacted by the explosion if AI slop. In fact, it seems like people are craving more of that honesty. However, if you spent your time as a creator just making low quality clickbait, you’re cooked.

That’s also true about being a Software Engineer. I think bad engineers are cooked. But only if they choose not to learn. These tools can help a bad engineer become a good engineer. It’s like having a someone to pair program with any time you sit down to write code. A good engineer will still invest in engineering knowledge. AI is a tool, if we must use it, we should use it to make ourselves better, not to replace ourselves.

I personally enjoy creating things and I’m not going to stop. I will not use AI on this blog, but I’ll probably be using at work and in other parts of my life. It makes my life easier. And yes, I’m conflicted about its impact, but it’s useful.

Idk.

• Wisdom