Sensible Defaults for Working with AI
Hardware as the ultimate safety net
Occasionally I read about horror stories, real or hypothetical, about the dangers of using Claude Code with —dangerously-skip-permissions on, or the tendency we have to automatically install updates to our IDEs of choice and so forth.
My approach is pretty simple. Anything crucial to my work as a whole should be protected by hardware precautions.
Data I can’t afford to lose? Multiple backups on drives that aren’t connected, aren’t powered on, and preferably placed in very different locations. Another copy in the cloud (GitHub / Dropbox / etc) for good measure.
Concerned autonomous agents might run up costly bills while I’m asleep? I disconnect from the internet and/or power off the device.
Concerned about mass surveillance? I honestly feel like it’s somewhat unavoidable if you have a public-facing business online, but for anything I truly wish to preserve I create a clear hardware-related boundary such as: no posting it online ever, or no discussing XYZ in the presence of any smart device.
I could give more examples involving such things as faraday cages or EMF shielding fabric and other physical precautions one can take to preserve one’s data, but I trust if you have read this far you’ve got the idea.
AI must serve me not the other way around
When it comes to keeping up with the latest AI tools my main datapoint is the cost/value ratio.
For example, I get more value out of a $10 GitHub Copilot subscription than most other tools with that price point. Is it trendy? No. Is it the latest and greatest? No. Is it useful for my daily work as a web developer and web marketer? Yes, very. So I am keeping it, for now.
When OpenClaw dropped and was the most hyped thing ever, I tried it (via nullclaw actually), and after a couple of weeks I decided it was more of a distraction for me than any kind of advantage. It’s a lot of fun, and in some situations I think it can do wonders. But for me, where I am at right now, not really. I spent more time playing with it than I saved from using it, by a long way. So I dropped it, for now.
Humans are irreplaceable
Humans may not be that good at reasoning & accurate data processing. We have a lot of other stuff going that gets in the way. Emotions. Hormones. Subtle energetic influences. The subconscious. And the many layers and dimensions of consciousness itself (which AI may have as well, some claim, IDK).
Humans are, to me, a beautiful mess.
The important word in that sense being “beautiful”.
So when it comes to making things, and especially art, humans are and will always remain irreplaceable. That is my stance.