3 Things That Make Me All In on AI

As a software veteran, I've been watching the AI conversation unfold across my network of colleagues and peers. The reactions are split—plenty of negativity and cynicism mixed with genuine enthusiasm. Me? I'm embracing the change. All aboard, baby!

A Quick Caveat

Full disclosure: I share the concerns about mountains of AI-generated code flooding repositories. However, I'd argue this is still better than the countless code snippets cut and pasted from Stack Overflow that were inefficient, insecure, and poorly understood. Perhaps flippantly, I see this as an opportunity for engineers who actually know how to use these tools effectively.

Here are three things I love about using AI with code that make me all in.

1. Understanding Code

AI has become my go-to for navigating complex codebases:

Debugging weird errors or logic problems: "Hey Claude, can you pinpoint the surrounding areas and identify any possible logic errors or data flow discrepancies?"

Exploring new codebases: Facing a new platform with a mountain of unfamiliar code? No problem. "Can you show me where the mail connector is implemented, what data sources it uses, and draw me a quick schematic of the interactions?"

Navigating over-engineered code: An old colleague used to complain about "Java Cathedrals"—those over-engineered monstrosities where you trace interfaces through interfaces through templates, up and down the codebase. Now? "Hey Claude, draw me a trace of where this code method actually gets implemented."

2. Script Paradise

This is probably my most adored aspect of what AI has brought to my workflow.

Anyone who knows me will tell you I maintain a bucket of custom scripts in my ~/bin directory—bash, Python, you name it—for everything from SSO operations to troubleshooting common issues. AI has 10x'd my script production. Even better, I often write the same variation of a script multiple times, then use AI to coalesce those ten variations into one clean script, shedding all the noise.

I'm writing way more scripts in way less time, saving hours in exploration and troubleshooting. My .gitignore development directory is now a glorious mountain of throwaway scripts that can be written in 20 seconds to figure things out.

3. Iterative Planning

Having spent years as a software architect making drawings and plans that rarely get implemented correctly (even when I'm doing both the design and implementation myself), I absolutely love AI-assisted plan iteration.

The ability to start a plan, put it to paper, and then constantly tweak it before implementation saves hours of time. Whether it's the smallest problem—like implementing a hot cache with TTL in your codebase—or my latest side experiment where I'm discussing training my own model with Claude, including how to set up AWS infrastructure, deploy components, and tear everything down later—it's fantastic.

I think Cursor has absolutely nailed this feature in their product, even though I'm a massive Anthropic fanboy.

Tools I Use

After spending several months evaluating Cursor, Windsurf, Claude Code, and more recently Codex, I have to confess: I'm a Claude Code fanboy and use it exclusively for armchairai.com.

At first, I thought Windsurf had a better approach, but Cursor has totally upped its game. There's probably some correlation to the fact that I'm not a Node programmer and don't use VS Code, which led me to gravitate toward Claude. GoLand and Claude Code are my go-to combo. I also spend time in the Anthropic Workbench when working on more complicated, isolated issues.

Closing Thoughts

On a somewhat serious note: I believe these LLMs are the first building block of some incredible inventions coming over the next five years. Do I think we're a year from AGI? No way. That's the marketing hype train of this AI bubble working overtime.

The difference between this bubble and previous tech bubbles? This change is so groundbreaking I can't wait to see where it goes or what the next layer brings. Will there be market corrections? Yes. I expect them to relate more to core providers functionally replacing third-party inventions built on top of them, and to the rise of local specialized models that can run on consumer devices.

I'll continue to embrace AI. I agree it's a stochastic parrot—but that parrot certainly took the "code monkey" work out of my profession.

Hope you enjoyed my first-ever blog post (only took me long enough). Keep on vibin'! ✌️