Go back in time a year and a half- it's mid-2024, LinkedIn has discovered AI and now the buzzword of the year is "agentic". Everyone and their brother was trying to convert every single task to be doable by an "agent" and, to be blunt, they sucked.
If you ever looked at my online posts or my LinkedIn, you'd have found that I simply didn't like agents. I think part of my original disdain for them came from this period. Tech "thought leaders" (I hate that term) were pushing agentic everything so hard that it made me feel like the entire concept was snake oil... especially since it seemed like most of the agents were failing at their jobs. The outputs looked terrible, folks were spending tons of time fixing what the agents were doing, etc etc. Thanks to this, I leaned very heavily into the "Workflows are life. Agents suck. I'll do everything by hand" mentality, just because I got so tired of the whole push to turn everything into an agent, even if it made no sense.
Well, times change I guess... because here I am in 2026, thoroughly enjoying Claude Code.
Now, I absolutely understand a lot of the current criticism in the dev world about agentic coding, but I feel like a lot of it really boils down something that you also see in AI image and video generation: most of the "slop" is because it was created with extremely low effort, resulting in something extremely low quality. Too many people just type the equivalent of "Masterpiece, best quality, anime girl poster that I can sell for money" and then declare it the greatest thing since sliced bread. Many vibe-coders do the same thing in code: "I need an app that does A, B and C" and then they let it run wild. These are basically the AI equivalent of sketching something on a napkin and calling it a commission.
I have even less sympathy if you're actually a developer or an artist working that way... at least have some pride in your work.
Note: I will give up development and take up farming before I ever refer to what I do as 'vibe coding'. Add that alongside 'Thought Leader' in Socg's "Upsetting Words" compendium.
For me, especially doing dev in languages I'm more comfortable with (like C# .NET), it gives me a chance to really step back and do more architecture than raw coding.
Case in point: right now I'm working on a small side project that I probably won't open source- something to run inside our home network do some work for us. It's a .NET 10 app using Blazor for the front end. It's nothing special, but I need some practice with it anyhow. For the past 3 days I've been spending my free time working with the claude chat on their website, doing deep researches and talking through design patterns, libraries, plans, etc to come up with a full dev plan before ever handing it off to Claude Code.
Not a single line of code has been written yet, and in fact the solution doesn't even exist yet, but so far the entire app structure, libraries, design patterns we're using, ORM we'll be using, DB tech we'll be using, deployment strategies, cost estimations for an eventual Azure deployment, etc have been talked through, investigated, and ultimately solidified.
The driving force behind all of these decisions is a combination of 2 things:
- My own knowledge and experience, plus any studying, videos, documentation, etc that I'm utilizing
- 13 or so individual Deep Researches, that are among 30+ individual chats with Claude, most of which result in documentation that I then collect and take to the next chat to talk through more stuff.
Yes, it can be grueling. But you know what? It's also fun. It's really fun to abstract the dev out to the architectural level and know that I can just delegate the actual efforting of making the final product to something else. As long as I've specified exactly how it all should be written, I can leave the final work of getting the code written, unit tests added, etc to Claude Code.
I've been a team lead and dev manager for most of my career- probably a combined 12 years of leading at this point. But this level of freedom over the decisions and the doing is something that I never had. Generally as a team lead or dev manager, architecture is either dictated from above you, or decided at the team level. Even if no one was dictating it to you- unless you want your devs leaving out of frustration or boredom, you don't get to dictate it yourself.
And the doing? For that you have scrum, agile, etc processes, run by PMs, to break down work and divvy it out to everyone. And that divvying process is absolutely not fun; that whole process is opposite of fun.
So this ability to be total dictator over the architecture and yet also be completely hands off from the more annoying elbow-grease parts of making the thing? Just dictating exactly how it should look, how it should work, etc and being able to micromanage the crap out of this little robot as it builds whatever I ask? It's all something I'd never do, nor want to do, with an actual team of people... but good lord is it so fun with AI.
Yea... this is enjoyable. I really like working like this. I'll admit that I'm feeling pretty nervous with some of my knowledge gaps... having other people around, on a team, really helps you feel more comfortable in knowing you're less likely to have done a dumb. But all the same- I'm enjoying this era of AI development.
Top comments (0)