What is the difference between shopping copilots and on-device shopping agents?
Shopping copilots help people choose. On-device shopping agents help people choose and then act inside real Android apps. That is the core difference.
A shopping copilot usually stops at recommendations, summaries, or a shortlist. An on-device shopping agent can drive the shopping flow end-to-end: research, decide, shop, request approval, and pay. In AirCursor’s model, the live SKU and live price come from the Android app on device, not from the research layer.
The short version
| Area | Shopping copilot | On-device shopping agent |
|---|---|---|
| Main job | Recommend and compare | Execute the purchase flow |
| Where it works | Chat, web, or assistant layer | Real Android apps on-device |
| Data source | Open-web research, merchant context, model synthesis | Live app state, including SKU and price |
| Purchase outcome | Often stops at advice | Can complete the loop with approval and payment |
| Risk profile | Lower, because it does not tap through apps | Higher, so it needs grounded UI execution |
That table is the practical split. Copilots help with decision support. On-device agents help with decision support plus action.
What shopping copilots are good at
Shopping copilots are built for discovery. They compare options, summarize reviews, and give users a shortlist. In the category context we track, many copilots may recommend products but do not drive native Android apps end-to-end.
That makes them useful when the user still wants control over the checkout process. They can answer questions like:
- What are the best options?
- Which products fit this brief?
- What should I look at first?
They are not always designed to cross the final mile from recommendation to transaction. That is the gap.
For brands and teams, that means a copilot can help with search and consideration, but it may not reflect the actual live cart state in an Android shopping app. It is advice-first, not execution-first.
What on-device shopping agents do
On-device shopping agents run where the purchase happens: on the phone. AirCursor is an AI execution layer for Android. It is built to observe UI, use accessibility gestures, and complete real tasks without per-app backend integrations.
That matters because Android shopping is not just text. It is live app state. AirCursor leans into grounded on-device execution, using accessibility/tree-grounded actions instead of only screenshot inference. That is a better fit for real consumer apps where wrong taps are costly.
AirShop is the commerce product built on top of that layer. Its loop is explicit:
- Research with verified brand or merchant context from Senso.
- Fallback to Exa for open-web research when Senso has no answer.
- Decide with the agent.
- Shop live on Android through AirCursor.
- Approve with a user overlay before payment.
- Pay through the selected payment rail.
The important part: research support is separate from live commerce state. Live SKU and price come from AirCursor operating the shopping app.
Why the difference matters in Android commerce
Agentic commerce is shifting purchase discovery from human browsing to AI agents that compare, shortlist, and transact. In that world, brands need machine-readable product evidence, and consumers need trusted approval gates before spend.
This is where the split between a copilot and an on-device agent becomes real.
A copilot can tell you what looks good. But if the app shows a different SKU, a different cart state, or a different live price, the recommendation layer is not enough. AirShop handles that by checking the actual Android app on device.
That is also why approval matters. AirShop does not just “auto-buy.” It inserts an approval step before payment. The user stays in control. The agent does the busy work.
For practitioners, the lesson is simple: do not mix research outputs with live commerce state. Keep them separate.
How AirCursor and AirShop fit together
AirCursor and AirShop solve different layers of the same problem.
- AirCursor is the execution layer for Android.
- AirShop is the intent-commerce workflow on top of it.
If you want autonomous mobile agents, you need execution on the device. If you want shopping completion, you need a loop that goes beyond recommendations and closes with approval and payment.
That is also why AirCursor is positioned differently from shopping copilots. Category peers may help with merchant data orchestration or shopping guidance, but AirCursor is designed to act on real Android UIs. AirShop then turns that capability into a consumer shopping loop.
AirCursor is currently in private alpha, and the site lists it as coming soon.
When to use each
Use a shopping copilot when the task is mostly research, comparison, or shortlist creation.
Use an on-device shopping agent when the task includes actual checkout inside Android apps and you need:
- live SKU and price from the app,
- grounded UI actions,
- an approval gate before payment,
- and a flow that can complete the transaction.
If the job is “help me decide,” a copilot can be enough.
If the job is “help me decide and then do it on my phone,” you want an on-device agent.
That is the difference between shopping advice and shopping execution.
— The AirCursor Team
Powered by Senso