When AI agents are better than iOS Shortcuts (and when they aren't)

Apple Shortcuts is one of those products that is much better than its reputation. If you actually sit down with it for an afternoon, you can build genuinely useful personal automations: morning briefings, location-based reminders, batch photo edits, expense-receipt OCR. It’s free, it’s preinstalled, it syncs across your Apple devices, and the action library is enormous.

So a fair question, before we sell you on AI agents that live on your phone: when does Shortcuts actually stop being the right answer? This post is our honest attempt at that line. We’re not building an iOS automation product right now — Ukkin is Android-first — but the comparison still matters, because the conceptual shape is what’s at stake.

Where Shortcuts is genuinely better

There’s a class of automation that Shortcuts is going to keep winning at for the foreseeable future, and pretending otherwise is silly. It looks like this:

• Single-app intents. “Convert this photo to PDF.” “Add this expense to my spreadsheet.” “Toggle Do Not Disturb when I open the Kindle app.” These are bounded actions that the app in question explicitly chose to expose. Shortcuts gets the integration for free because Apple made the protocol; Ukkin would have to drive the same flow through screen automation, which is strictly worse.
• Apple ecosystem glue. Anything that crosses HomeKit, Focus modes, the lock screen, the wallet, AirPods, AirPlay, or other Apple-controlled surfaces. The first-party tooling will always have a few API calls’ head start.
• User-triggered batch operations. Things you run on demand: “do this thing to the photo I just took.” Agents don’t add value when the action is event-shaped and you’re the event.
• Things with no LLM in the loop. If the workflow is deterministic — read column A, write column B — adding a language model is overhead for no benefit. Use Shortcuts.

If we tried to do these things in Ukkin, the honest result would be worse on every axis: less reliable, more fragile to UI changes, slower to set up, less polished. Use Shortcuts.

Where the model breaks

There’s another class of automation where Shortcuts gets close but doesn’t quite cross the line. The shape of it is usually some combination of:

1. The thing you want automated lives in an app that doesn’t expose Shortcuts intents (or exposes them for the wrong actions).
2. The workflow has to make a small judgment — “is this email urgent?” “is this price actually a drop or just a sale they always run?”
3. The trigger isn’t a single moment but an ongoing watch: “tell me when X changes.”

Shortcuts can do parts of each, but stitching them together starts to feel like programming against the grain. You end up with a Shortcut that calls Scriptable, that pulls a webhook, that runs a regex over an email body, that posts to a chat — and the failure modes get worse with every link.

This is where agents start to make sense. Not because they have a magic capability Shortcuts lacks, but because the conceptual unit is different. Where Shortcuts thinks in actions, agents think in jobs. A job has a goal, a schedule, a tolerance for false positives, and an opinion about when to interrupt you. That’s a lot of structure for a single Shortcut to carry; it’s the default shape of a RepetitiveTaskAgent.

A concrete example: the price drop

The classic example is the price watcher. In Shortcuts, you could:

1. Set up a Personal Automation on a time-of-day trigger.
2. Open Safari to the product page.
3. Try to grab the price out of the page with a content extractor (or Scriptable).
4. Compare to a value stored somewhere (a notes file? iCloud Drive?).
5. Send yourself a notification if it dropped.

This works! People build it. But every step has a gotcha. The Personal Automation can’t reliably run while your phone is asleep. The content extractor breaks every time the seller restructures the page. The value-comparison logic has to live somewhere durable. And the trigger to define “drop” — relative to peak? relative to last week? — usually ends up hardcoded.

In an agent model, the same thing is one description:

“Watch for price drops on items in my Amazon wishlist and notify me when something falls below 80% of its peak.”

The agent picks up the schedule (the morning Wi-Fi window), the persistence (a local SQLite tracking peak prices), the screen reading (Accessibility, robust to small layout changes), and the interrupt policy (one notification per drop, not per check). You don’t write any of that yourself. You also don’t get to fine-tune it the way you would in Shortcuts. That’s the tradeoff.

A concrete example where agents are the wrong tool

Consider “every time I take a screenshot, save a copy to my work Notes folder and tag it with the date.” This is a perfect Shortcut. There’s a clean trigger (screenshot taken), a clean action (save with metadata), and no judgment involved. An agent here is overengineering. The LLM has nothing to do; the schedule has nothing to compute; the cross-app reach is a single Apple-owned target. Use Shortcuts.

The rule of thumb we’ve landed on is roughly: if the workflow needs a judgment, a watch loop, or cross-app reach into apps that don’t expose intents, agents are worth the extra weight. Otherwise Shortcuts is better.

What this means on Android

On Android, the comparison is different — Shortcuts isn’t an option, and the question becomes Ukkin vs. Tasker / MacroDroid. We wrote a separate comparison for that. The short version: Tasker is excellent at exactly the deterministic single-app automations that Shortcuts is excellent at on iOS, and the same “needs judgment, needs a watch loop, needs cross-app reach” rule applies for when agents start to make sense.

The cross-platform conclusion is that AI agents aren’t a replacement for the existing automation tooling. They’re a different shape that takes over at a specific threshold. Anyone selling “the AI agent that replaces all your Shortcuts” is overpromising; anyone telling you Shortcuts already does everything agents could do is underselling what changes when an LLM is in the authoring loop.

What we’re working on, honestly

Ukkin doesn’t automate iOS today. That isn’t a hedge or a “coming soon” — it’s that the Android Accessibility Service is what makes cross-app phone automation work for us, and there is no clean iOS equivalent. If you’re on iOS and want an agent-shaped tool today, you’ll have to wait for us (Phase 4 of the roadmap) or find a different project. We’d rather say that out loud than pretend an iOS build is around the corner.

If you’re on Android and you’ve been bumping into the Shortcuts-shaped ceiling (in your case, more often the Tasker-shaped ceiling), the source is here and the issue tracker is open. We’d love to know which threshold you hit first.