Coding on the Fly
The ups and downs of coding on the fly (literally)
tl;dr: AI tools have provided a huge improvement in the ability to develop software for mobile devices, but they’ve added new cracks in the pipeline to be aware of.
They say the best camera is the one that you have with you, and that’s true of computers as well. I’ve been using an iPad of one sort or another as my primary mobile computer for nearly fifteen years, tackling more and more work while striving for the holy grail — software development. That’s always been the most fraught and fragile use case for me. AI has made this both better and worse.
AI development tools have dramatically improved mobile software development, making it super easy to tell the robots what I want them to do while I’m on the go. I can start sessions on my desktop and continue from my phone while I’m in transit or in a lull during one of those away-from-keyboard moments. Claude will even send me notifications when it needs my attention, so it’s easy to give a short answer in the right place to keep things moving along. It’s very reminiscent of talking with folks while walking between meetings back in my more corporate days, using those small snippets of time to get things done. It’s a well-practiced skillset and it’s definitely made me more productive while having a better work-life balance lately.
At the same time, all of the conditions have to be just right, else my productivity comes to a screeching halt. Are all the right Claude Code sessions available with remote control turned on? Is my Claude desktop app open so that all my scheduled tasks will run and Cowork Dispatch will connect with it? Is Tailscale in good shape so I can ssh in from my iPad? Docker? My Chrome Sandbox for Claude with all the right things signed in? Are all my MCP servers properly authenticated? Have I recently entered my passphrase so commits will be properly signed? The list seems nearly endless.
Here’s a quick example from my life: I had a quick one-night trip to visit a client mid-week, hopping on the big purple bus late morning on a Wednesday, returning Thursday evening. I had some development work planned for the flight out, intending to use both SSH and remote control to make productive use of the flight time. It wasn’t until we were free to move about the cabin that I realized Tailscale was down on my desktop, and my remote control sessions needed various forms of action directly on the box in order to continue. Why was Tailscale down? I don’t know, but that simple thing that I could have checked before I left killed my intended productivity for that block of time. It lessened the sting when the airplane wifi had significant latency and dropouts, though; by that time I had pretty much given up.
Still, I don’t want to come off as the old man yelling at clouds. It is phenomenal that I have in my pocket an amazingly powerful computer that can tell amazingly powerful robots to go do useful things, and they get done. If I think about the distance from where we are now back to where my personal journey in computing started, it’s as mind-blowing as realizing that smartphones aren’t yet 20 years old. And they’re getting better, and they’ll keep getting better. They just take careful understanding and some babysitting right now, which has always been true of nearly any new technology until we have used it and understood it for so long and so well that it becomes nearly invisible … kinda like the internet you’re using right now, probably on your smartphone.