Thoughts About AI from a Tech Conference

I’ve just finished Day 2 of the Enterprise Tech Leadership Summit in Las Vegas. I initially wanted to attend to get some ideas to bring back to one of my teams at work to help improve their development workflow. The conference is organized by Gene Kim, author of The Phoenix Project and The DevOps Handbook, so I expected lots of tips about DevOps best practices. It turns out DevOps is basically just table stakes at this point. Not surprising, but the majority of the conference was focused on AI.

AI was EVERYWHERE at this conference. It’s easy to see AI as the latest hype-cycle cult thing in tech, like crypto/blockchain before it, or even Agile or DevOps. But man, it really feels like there’s something there. Both during the talks and in informal conversations, seasoned, well-respected developers say they now have superpowers because of AI. Never learned Ruby but wanted to update a web service you support? No problem—just vibe-code it!

Vibe coding in particular seems to be having a moment. Former developers turned middle managers, and even product and design people who’ve never coded (but understand software development) are making their own web apps and having a blast. They have stories of staying up all night “coding” like they did in their 20s.

The most impactful AI tools I’ve seen yet are the terminal/command-line coding “agents.” Claude was one of the first with Claude Code, but Google and OpenAI have their own versions too. Instead of pasting code into a chat window and having it rewritten, or pointing a chat client at your IDE, these tools live in your terminal and can be given access to create, write, delete, and move files (be careful!). They can create a git repo, commit changes, search the web, and stand up web servers when needed. You can run them in supervised mode, where they create suggestions you must approve, or what many call “YOLO” mode.

I’ve been experimenting with Codex, OpenAI’s tool, and tried Google’s GeminiCLI today at a Google talk. I was able to update this site’s design with it and vibe-coded a little QR code generator app in about five minutes at the Google CLI workshop. It’s true what they said—it’s fun and it does feel like you have superpowers! Many devs talked about having 10–12 terminal windows open at once, acting sort of like engineering managers: give instructions to one, let it work, move to another, then check back.

I’ve even been playing with the command-line tools in my Obsidian vault. It’s a great use case since Obsidian vaults are just collections of plain text files. So far I’m experimenting with having an agent read the contents of notes created at the root and suggest where in my directory structure to file them. I’m having it create a Markdown file with the suggestions for now, but if it works well I might YOLO it and make a small autonomous sorting agent on my Mac Mini to organize my vault automatically.

Anyway—definitely having fun, and it definitely feels like there’s something to this AI stuff. (Or maybe I’ve been brainwashed by the AI cult? Just as likely.)