A real-world experiment in using AI agents, Tailscale, Apple signing, and a private catalog to build and install iPhone apps while away from home.
I was working on a few features when I had to leave the house for an event.
Once I arrived, I found myself bored and wanted to keep going. My laptop was still at home, so I used remote functionality to talk to it and continue building.
The agents could build the feature. But I still needed a way to test it.
That became the real problem. Without being on my local network, I had no convenient way to get the latest build onto my phone and verify that the feature actually worked.
Could an agent build an iPhone app on my Mac, sign it locally, and deliver it to my phone over a private network?
My first instinct was that Tailscale might make this possible. It could provide encrypted connectivity between my phone and laptop, even when I was away from home.
The basic idea seemed straightforward:
I asked Fable 5 to figure out how to make it work.
After thinking for a while, it gave me a flat answer: this was not possible and was not allowed.
That answer was disappointing, but it did not feel complete. I understood that Apple’s signing and installation rules mattered. Still, I wondered whether the proposed workflow was being rejected too broadly.
If it works over my home network, why should a private Tailscale network make it impossible?
Tailscale would not bypass Apple’s requirements. It would only provide the connection between devices. The app would still need to be built, signed, and installed through a valid process.
I gave the task to GPT 5.6 Sol.
It thought through the problem for roughly 15 to 20 minutes. The result surprised me. Instead of stopping at a general restriction, it worked through the individual pieces of the workflow and found a path that matched what I was trying to do.
The solution was to have my Mac host a small private catalog. My phone could reach that catalog through Tailscale, creating encrypted connectivity between the two devices.
The network made the catalog reachable from almost anywhere, while the Apple signing process remained the mechanism that established whether the app could be installed.
The agent automated the sequence I would normally perform by hand:
When I tap the link, the phone requests the manifest and the IPA from the catalog.
Apple verifies the app’s signing information, and the device proceeds with installation when the build and signing setup are valid. The important distinction is that Tailscale handles reachability, while Apple’s signing process handles trust and installation eligibility.
The final experience is remarkably simple:
The agent builds the app, signs it, publishes it, and gives me a link. I tap the link, install the app, and test it.
The complexity stays on the Mac. The interaction on the phone is reduced to opening the install link and waiting for the build to arrive.
This changes the meaning of remote development.
It is not only about controlling a computer from a distance. The more useful loop is:
The breakthrough was not a mysterious workaround. It came from separating the problem into clear parts.
There are still practical boundaries.
The Mac must be available, the signing setup must be valid, the phone must be able to reach the private network, and the catalog should not be exposed carelessly. A private network improves access control, but it does not remove the need to protect certificates, devices, and installation links.
One agent said the goal could not be done. Another kept working through the details until it found a viable architecture.
The difference was not simply intelligence. It was the willingness to decompose the task, distinguish restrictions from assumptions, and test whether each part could be connected without violating the underlying requirements.
Sometimes the useful answer is hidden behind an overly broad “you can’t.”
The better question is often: Which exact part is impossible, and which parts are merely inconvenient?
In this case, remote access, private networking, local signing, hosting, and device installation could be combined into a workflow that let me keep building from anywhere.
Whether I am at home, 100 miles away, or 1,000 miles away, the workflow is the same.
The agent continues the implementation. My Mac produces the signed build. Tailscale provides the private path. My phone receives the app, and I can test the feature on real hardware instead of waiting until I return home.
Before this experiment, remote development stopped at the laptop. After it, the process reached all the way to the device in my hand.
That is what made the result feel so powerful: the distance between an idea and a testable build became much smaller.