Vibin'

Introduction

At this point most of us devs have been beaten down by the nonstop hype and hate around AI tools (primarily dev tools). Depending on who is talking, these things are either going to 10x our productivity, put us all out of work before lunch, or are so worthless we should not touch them at all. A lot of that noise is complete bullshit. Non-technical C-Suiters, VCs, and sales types love to hype this stuff because there is almost always a financial angle, or they are just riding the latest hype train (blockchain flashbacks, anyone?). On the other end of the spectrum, the people who dismiss the tools outright are usually reacting more to their own hangups than to the tools themselves.

Like a lot of folks in the industry, I am skeptical of anything that promises magical productivity boosts. My team has also been asked to increase our use of AI tools, which means I have had plenty of chances to mess with them in real work. This post is about how I actually use them, where they help, where they fall flat, and how I sift through all the hype and doom. It is all anecdotal and based on my own experience. I am not an AI expert. I am just a dev who has spent enough time with these tools to have opinions.

Hypers

Every day on my LinkedIn feed, on certain news channels, and across the internet in general, I see the same AI hype posts from some CxO, VC, or other suit talking about how AI is going to render us obsolete or 100x our productivity. Then there are the endless announcements about a new company using AI to revolutionize X (just do a find + replace with using blockchain to revolutionize X … same energy).

This isn’t an original thought, as many have said it before me, but: I genuinely believe that most of this hype exists for financial reasons, and I can’t take anyone who plays into shit in this manner seriously.

Posts telling you to get on the AI train before it’s too late are trying to get you using their products so they can make money off you. Companies bragging about how they are ReVoLuTiOnIzInG something with AI are usually doing it to check a box for VCs. The guests on these news shows are often investors or company reps with a direct or indirect financial interest in the success of the technology. All of it blends together after a while, and it’s exhausting.

Haters

On the other end of the spectrum are the devs who refuse to use AI tools and / or will not put in the effort to learn how to use them productively, then dismiss the whole thing as garbage. I am not talking about people with valid environmental or ethical concerns here; a lot of those I actually agree with. No, I am talking about something else entirely.

I’ve had conversations with other devs in person and online, lurked on various threads, read people’s blogs, and I’ve formed a fairly consistent opinion: some people just don’t want to use these tools regardless of how good or bad they are. In some cases, instead of saying that and moving on, they come up with all sorts of negative shit to try and justify it (there are plenty of lazy anti-AI tool rants out there, not going to restate them here).

Why? Honestly, I think there are a bunch of reasons, but one stands out: programmers typically like to program. For some people, their love for the craft fuels their reluctance to use AI code generation tools. One of the joys of programming is encountering interesting problems and building your own working solutions. There is a real sense of satisfaction there, and it can feel like you are being robbed of that if the LLM is doing most of the heavy lifting.

How I Use AI

I’ve been using AI in various ways for fun since the OpenAI beta, and more recently I have been using the technology daily at work. When I first got into the beta, my reaction was basically hey, neat toy and I used it to answer simple questions and generate some funny stories. Coding with it initially sucked.

The landscape has changed a lot since then – tons of AI coding tools out there, more competing LLMs, and a handful of services that make integrating with LLMs pretty painless.

Ok, so how am I actually using this stuff?

Searching Online

First and foremost, Google search is unusable for me at this point (and their AI-generated answers at the top are a joke lol). If I use a search engine at all, it’s DuckDuckGo, but a lot of the time I am taking advantage of the fact that I can just ask ChatGPT and it will respond with an answer and links I can click into. The quality of the responses has gotten noticeably better, so it’s hard for me not to use it.

Non-Coding Work

I have my primary LLM tool hooked up to various MCP servers, which I use to write and iterate on things like ERDs, tickets, and other non-code-y tasks. I encounter a lot more of this sort of work now at this stage in my career, and I am not going to lie: I still hate it.

Previously, ERDs took me longer than they should have because I loathe writing them. With an AI tool wired into the MCP server for our knowledge base, I can just have a conversation with the LLM, point it to relevant code, docs, and tickets, and “write” the document section by section. Similarly, I hate writing tickets … but AI helps here too! I still have to read and correct the output, but the back-and-forth makes the process suck a lot less.

Un-Sexy, Boring Tasks

One of the biggest killers of motivation for me is work that is easy, quick, and / or completely uninteresting. Historically, I would put these tasks off, feel annoyed about doing them, move slower than I should, and then get stressed out because they piled up. Meanwhile I would gravitate to the interesting stuff, letting the mundane mountain grow.

AI tools have helped me a lot here. This category of work is basically perfect for an AI agent, and the agent is often faster than I would be even on my best day. On a recent project involving a migration between two repos, the entire thing was 90% boring grunt work, moving and renaming files. I generated an ERD and a set of tickets (using the conversational flow I mentioned above), then handed each one to a coding agent and let it run.

The whole project took a fraction of the time it would have taken manually. I found myself reviewing code and making only minor changes (and since it was not a greenfield effort, there was not much for the AI to screw up anyway). The agents also handled linter issues, test breakages, and other cleanup. This all allowed me to shift my focus to other things as needed without being so stressed about it … honestly felt like a win.

Writing, Generally

I am not much of a writer. I mostly write the way I talk, which people either love or hate. For this blog I try to clean it up a little, so I have given ChatGPT a set of instructions (after a lot of trial and error) to help keep things in my voice while making the writing readable.

I take it section by section and usually feed the tool a pile of run-on sentences, 5000 swear words, and a dozen tangents. What comes back is something properly punctuated, with about 85 percent fewer swear words and most of the tangents removed. Hate on it if you want, but the biggest thing that has kept me from blogging for years is that I hate the way I write. The AI assistance makes me hate it slightly less.

What I Don’t Like

There is a bunch that AI tooling is great at. There is also a lot it’s still ass at, and for everything else I’ve just decided I don’t want to use it at this point.

Using It For Complex Work

AI does not handle complexity well, and the problem gets worse the longer your session runs and the more context you shove through it.

I had a project that was legitimately complicated and required a lot of context about code, business expectations, and interactions between different systems. I said fuck it, let’s try to vibe this, shared an ERD, multiple tickets, various docs, and some tribal knowledge into the session, and hoped for the best. The result was a huge mess and me wanting to throw my laptop. And of course I blew through the context window almost immediately.

I did not abandon AI entirely for the remainder of the project, but I had to be way more surgical than I planned, and the initial thrashing was time wasted.

Spiral Debugging

Sometimes you point an LLM at a test failure and it fixes it the right way, no problem. Other times, it tries to mock out the universe and rewrite half the test suite so that 1 == 3 passes. Similarly, sometimes it can debug an issue, other times it just throws shit at the wall hoping something will stick.

I am at the point where I can tell pretty quickly when I am going to have to step in and tell the AI to sit down and shut up, but it still irritates me to no end. In the past I tried thrashing with no, look at <x> only for the tool to agree with me and then immediately do something else wrong. Big lesson: learn to cut your losses quickly.

Vibe Coding In Personal Projects

I don’t use AI coding tools in my personal projects, at least not yet. Even though I use them daily at work and get real productivity out of them, that is not the point of my personal projects. When I am building something on the side, I want to have fun and usually learn something new, not crank out tasks faster.

I’m not ruling out using AI for personal stuff in the future, but right now I have my programming life split into two boxes: my personal box where I code by hand, and my professional box where I code by hand and vibe. So far, I have been happier keeping that separation.

Closing Thoughts

AI tools are useful means to some ends. I don’t believe they are going to steal my job, no matter how often we hear that line, and I don’t think they are so utterly useless that you should ignore them. I don’t love any of these tools so much as I simply use them because they are usually good enough. I use OpenAI’s and Anthropic’s tools a lot, but I am by no means a fanboy of either.

My general advice is this: don’t be so starry-eyed that you forget your responsibility to think, and don’t be so negative that you dismiss an entire category of tools. In my experience, there are clear productivity gains once you learn your way around. On the other hand, you will also learn when it’s time to shut down the session and blow the dust off your text editor ;)

There is a lot more I could say, but we will leave it here for now. Thanks for stopping by!