AI assistants and MCP: what changes for mobile.

The easiest way to picture the change is the moment you'd actually see. You type one sentence to Claude or ChatGPT or whichever assistant you talk to. It comes back with a confirmed plan. No app to open. No carrier portal. No QR you have to keep in a screenshot.

That exchange used to be an app install, a country picker, a plan comparison, a checkout, a saved card, a QR code on a different device, and a system settings screen. Now it's one sentence.

But the visible part is the smaller part. The bigger shift is everything the agent does without being asked: the top-up scheduled before you run out, the activation retried when it hangs, the carrier swap surfaced when a cheaper plan appears, the line switched off when the trip ends. The chat is where the agent talks to you when there's a call to make. Most of the work happens before that.

Travel is where this lands first. Anyone who flies regularly has lived this loop: remember the trip on Monday, forget to add data, land on Thursday with no signal, lose the first hour looking for WiFi to set something up. The usual outcome is a plan you added on the runway because you finally remembered while taxiing.

With an agent, you don't remember. The agent does. It sees the flight in your calendar, picks a plan against the policy you set ("anything under $10 in Japan, just do it"), and either provisions the line directly or pings you for the call. The same loop closes for the longer-stay traveller: the agent notices you've crossed into a new country, that the current plan doesn't cover it, and that the data is about to start charging at roaming rates. It switches the line before you've noticed there's a problem.

Travel is the obvious case. The bigger ongoing one is the quiet admin around your line that you should do but don't. Topping up before your data runs out. Renewing the plan for the country you keep visiting. Switching the line off when the trip is over so it doesn't auto-renew. Moving the line to a new phone when you upgrade.

An agent with access to your line handles these in the background, on a schedule. The top-up runs the day before your data would run out. The line gets switched off the morning after your trip ends. The number moves to the new phone the moment the agent sees the new device on your account. The notification only lands when there's a decision you have to make. The work that used to happen at 2am in a hotel lobby happens at no time at all, because no one had to do it.

The classic mobile failure mode is the silent one. The phone shows bars but the data won't load. The eSIM is on the device but never activated. The plan is active but in the wrong country. The usual fix is a call-centre, on hold, while the trip moves on.

The agent doesn't wait for you to notice. It watches the line's status. If activation didn't complete, it refreshes the service and pushes the profile again. If the plan ran out three hours ago, it tops up before you've opened the settings screen. If the phone is in a country the current plan doesn't cover, it provisions a local option as you land. The fallback is still there: you can flag the problem and ask the agent to look. But for most cases that used to mean a call-centre, the diagnosis and the fix have already happened by the time you'd have picked up the phone.

Most people stay on their mobile plan because changing it is more work than the saving is worth. The carrier knows this. So does the agent.

An agent watching your usage on a continuous basis does the thing you'd never get around to. It pulls the latest plan catalogue, compares it against your last three months, and surfaces a switch when the numbers move enough to bother. "You used 4 GB this month across three countries, the next tier up saves you about $7 a month." "You've spent $43 on roaming in the last quarter, a local eSIM for those trips would have been $11." "This carrier dropped their EU plan price last week, switching saves you $4 a month at your current usage." That's a conversation the carrier has never been able to have with you, because it isn't watching the way an agent watches.

Adding lines works the same way. A second line for the new phone, a dedicated line for travel, a separate line for a kid's device. The friction of provisioning multiple lines used to mean most people stuck with one. The agent provisions on request, or on a recurring policy ("a fresh travel line every time I book a flight outside the EU"), and the second line stops being a project.

For an IT team, this is bigger. The agent watches the HR system. The day a new hire's start date arrives, the line gets provisioned and the install link goes out, without anyone in IT touching it. The day someone leaves, the line is suspended and the data terminated, on the same trigger. Unplanned trips work the same way: the agent watches the calendar or the corporate travel system, sees the booking, provisions a local plan, and confirms it in a Slack message before the employee has even thought about it.

The helpdesk category for "I'm in a hotel in Singapore and can't connect, can you help?" stops generating tickets, because the activation already happened. The cost saving is real, but the bigger gain is that the IT team stops being the bottleneck for a recurring class of "small thing that takes three people to fix". The line, the activation, the top-up, the reclamation: none of them require a human in the loop unless the policy says they should.

For a deeper dive on the workflow side, see corporate roaming costs: the bill is not the real problem.

A pattern shift further out: the agent isn't always a general-purpose chat assistant. Sometimes it's the device itself. A car with embedded connectivity can request a new plan when crossing into a country its current carrier doesn't cover, on its own, without the driver doing anything. A wearable can request a small data top-up when its sync queue backs up. An industrial sensor can hand over to a different carrier when the primary network has been silent for thirty minutes.

That's the bigger version of this story. Connectivity stops being something a human picks for a device, and starts being something a device picks for itself, against policies its owner has set. The eSIM made the SIM digital; the agent makes the choice automatic.

Why is mobile connectivity early to this? Two reasons. The unit of inventory (a SIM profile) is already digital. The purchase flow is small enough that an agent can complete it without a human in the loop and without a lawyer in the room.

Identity, payments, logistics, energy, and healthcare are next. Each of them has a unit of inventory or a state change that an agent will eventually be allowed to handle. The providers that ship an agent-addressable surface first land in the agent's tool list. They become the default that the assistant reaches for. The ones that don't get there for a year or two get described to the user as "another option that's not yet available in this assistant".

For the broader embed-in-your-app angle, see eSIM for apps and the developer docs. That's our path to being one of the surfaces an assistant reaches for, not just a website you visit.

The agent experience isn't fully comfortable yet. Users don't always trust an AI to spend on their behalf, and they shouldn't blindly. Asking for explicit confirmation on every purchase makes the flow feel slower than the website it was meant to replace. Letting the agent act without asking makes the user nervous. The right pattern is a policy: budgets, plan ceilings, country lists, frequency limits. Above the line, the agent acts. Below it, the agent asks. Most products are still calibrating where to draw that line.

The second thing is visibility. When the agent acts in the background, you don't see it happen in real time. The top-up that ran overnight, the line switch when you landed in Madrid, the renewal that auto-paid yesterday: they happened, but you weren't there for any of them. The agent has to show its work somewhere you'll actually look. A timeline of what was done, on which policy, and what's queued. Without that, the experience tips from "useful" into "I'm not sure what just got charged".

And there's a quieter shift underneath. Agents pick from what they can reach. A carrier that hasn't shipped an agent-addressable surface doesn't turn up in the comparison. It isn't blocked, it just isn't in the room. The customer never hears about it. That's a real change in how consumer choice works, and the vendors realising it haven't all moved yet.

Two takeaways for anyone designing a product or evaluating one. First, make your service available to agents. The standard for that is Model Context Protocol (MCP), and it's increasingly the way assistants discover what they can do for a user. Second, pick service providers that have already shipped an agent-addressable surface. Connectivity is one of them.

The logic underneath both: the agent is becoming a distribution channel. Vendors that get into it first become the defaults the assistant reaches for. The ones that don't get described to the user as something they might find elsewhere.

The agent space is moving fast. Patterns, etiquette, and the division of labour between user, assistant, and provider will look different again in twelve months. Last verified May 2026.