I now see how it can be addictive: A mindblowing experience and feeling like I see the world a bit differently. A feeling of being super smart and powerful. Getting lured in with free samples then everything costs. How it grabs my attention when using it, and then when I am not using it, I yearn to get back in.
Is it productive to think of generative AI as a new drug?
I have been thinking of a project in the decentralized web arena and trying to find someone to hire or inspire to try it out. Then a former Archivist gave a workshop at the Internet Archive showing how Slack programmers use Claude Code and it seemed too good to be true. So I thought I would try it. I dove in with a $20/month Claude Code subscription and started making astounding progress in an hour or so. It came up with a plan and started to execute it; we swerved, changed, debugged. I soon ran out of tokens and I needed to pay for more, this time $100/month– ok, lets try it. Point 1: first bit is free, then the cost escalates.
It flattered me at every step– “Good idea” “That will make a much better UX experience” etc. I was super good at this, apparently. It said so. All the time.
For 3 years, I have used generative AI tools like ChatGPT and Google AI summeries for one-shot help on spreadsheet formulas, command lines, helpful research, occasional entertaining poems or pictures. But yesterday I thought I would take out Claude code to see if I could prototype a project I have been dreaming of: a simple downloadable self-hosted web server that could be used without any of the hassles and costs of domain name registration, https certificate registration, hosting costs. A web server for a decentralized world. It is a project that would take a full time person maybe a year to get right and propagated, and that person is hard to find and expensive. Someone called the project “Onion Press.” Could I vibe code it?
I am a computer programmer but out of date. This tool magically could configure all the newest packages, knew how to take suggestions really well and iterated and fixed its own bugs. It was like having an impossibly-fast programming guru right there, always smiling.
My mind was blown watching a machine that could take vague suggestions and render them in code– and fast! It is surreal to watch the machine go through iterations debugging the setup– reading logs from multiple components watching things have issues and taking corrective action. It knows much more about these components than I do, and acts successfully. Mindblowing. I am trying to understand the implications of such a new capability being out in the world.
I felt super powerful– maybe now I could make my dreams come true. I even asked chatGPT for a logo for the project with a brief description, and out popped a nifty logo and related icons. In seconds.
After yesterday’s binge, I have been thinking of improvements all day. I can not wait to get back to the keyboard. It all seems too good to be true. Longing for another session of feeling empowered. And when I could not get back in, I had a feeling of loss, like missing an imaginary friend:
My description here could be someone describing a first psychedelic drug experience, complete with an urge to tell people of my experience.
Then it occurred to me: Generative AI is a New Drug. And it doesn’t spill. Yet.