Three pivots in three months, one belief that didn't move

First death

In early January we built a thing that turned natural language into iOS Shortcuts. You'd say a sentence, and we'd compose 590+ native capabilities — alarms, Health, Calendar, Reminders, Files, Photos — into a flow that ran on your phone. The mobile n8n. Programming with your mouth.

We were excited for about a week.

Then we discovered Shortcuts can't stay resident in the foreground. Any non-trivial flow gets killed mid-run. A user would ask, "summarise my sleep last night into a Note every morning," and the entire pipeline existed — except iOS terminates background execution after 30 seconds, so it never actually ran twice.

The product worked in demos. It didn't work in lives.

Second death

Once we stepped back, the diagnosis was uglier than the bug: the entire experience was built on a runtime we didn't own and that could be tightened at any moment. Fragile by construction.

So last month we pivoted. No more automation flows. Mini-apps instead. Say a sentence, AI writes a full small app, the app runs on your phone. Same target audience, different shape. We were positioning ourselves next to Rork.

We even drafted the Twitter thread. "AI lets anyone create software." It was good.

Then Apple moved.

Rork stopped operating on iOS. a0.dev gave up. Wabi and Vibe Code have been blocked from updating for a month. Replit — a nine-billion-dollar company — was forced to make its generated apps open in an external browser.

App Store Review Guideline 2.5.2: "Apps should be self-contained ... and should not download, install, or execute code which introduces or changes features or functionality of the app."

Our entire architecture was the exact thing that rule was written to stop.

The irony — that same week, Apple shipped OpenAI and Anthropic coding assistants inside Xcode 26.3. Vibe coding inside Apple's own tool is innovation. Vibe coding inside a third-party app is non-compliant.

The rule is simple: on iOS you can build, but you can't run what you've built.

From tool to platform

I wrote in our group chat: "Maybe building apps on iPhone is just going to be history."

Three months, two complete rewrites. The first one we shipped, but the system wouldn't let it run. The second one we shipped, but Apple wouldn't let it ship. Both times we proved the technical thing, then learned the technical thing wasn't ours to ship.

This time around the lesson finally clicked.

For three months we kept asking the wrong question: how do we let users create software on iOS? And every answer ran into the same wall — the thing they create only lives on their phone. A creation no one else sees is a hobby, and hobbies don't compound.

The right question is: what do users create for, and who sees it?

Instagram started as a filter app. You took a photo, you slapped a filter on it, you saved it to your camera roll. Then they added a feed. Suddenly the filter wasn't the product — it was the create button for a different product. The feed was the engine. The tool stayed the same; the surface area changed everything.

TikTok did the same trick. Edit tool plus algorithm plus full-screen feed, fused into one app. Creation and distribution shipped together.

Tools and platforms differ on one axis: who owns distribution. A tool's value lives in its features, and someone else controls whether your features run. A platform's value lives in its distribution, and that part you own.

AI generation + full-screen feed + recommendation. What users make isn't an app. It's content.

Where we go from here

We're cutting every dependency on someone else's runtime. We're going back to the thing we believed on day one: ordinary people should be able to make things with words, and the things they make should have an audience.

Three months, three directions, same north star — let ordinary people create software. The shift is in the second half of that sentence. We used to think the goal was everyone owns their own tools. We now think it's everyone's creations have viewers.

— 2026.03.20