At its annual developer conference, WWDC26 (Worldwide Developers Conference 2026), Apple announced a new set of AI tools designed to radically simplify the integration of intelligent features into apps. Developers now have access to the Foundation Models Framework for embedding Apple’s proprietary models, along with an upgraded App Intents for intelligent app control through Siri.
Special attention at the event went to how artificial intelligence is reshaping the future of mobile apps. To discuss how these innovations will affect the development of real products with audiences in the tens of millions, we spoke with Viktor Gordienko, Senior iOS Engineer at inDrive, a global mobility and delivery platform used in 48 countries and more than 1,200 cities worldwide. Before that, he worked on digital products at amoCRM, one of the leading sales and customer relationship management platforms, and joined the European startup Take It Personally as the second iOS developer, helping build the app almost from the ground up. Today, Viktor is exploring how AI is changing not just apps themselves but the entire approach to building them, from architecture and testing to the developer’s role on a product team.
In this interview, he explains how AI agents in apps can operate without sacrificing control over user data, and why that’s become one of the key challenges for developers.
– Apple claims at WWDC26 that their new AI tools mark a breakthrough for the whole industry. Viktor, as an iOS developer with nearly ten years of experience, do you think this is genuinely a new stage, or have we seen announcements like this before?
Apple has regularly introduced new technologies, but in most cases they extended the existing capabilities of devices or apps. With Apple Intelligence and App Intents, it’s different: the whole model of how users interact with mobile services is changing. In the past, an app was a set of functions that a person had to find and launch on their own. Now it’s gradually becoming an intelligent assistant that understands context, takes the user’s habits into account, and can carry out part of the work without the user’s direct involvement.
That’s why this shift is getting so much attention. In my view, we really are at the start of a new stage in the evolution of mobile apps, assuming Apple manages to deliver on everything it has announced. And we’ll likely feel these changes within the next few years: not just what the services themselves look like, but how we interact with them. Smartphones will probably stay with us, but their role will shift. They’ll become the main interface for interacting with AI agents that increasingly act on the user’s behalf.
– Can you give an example of this kind of interaction, so it’s easier to picture what will actually change for people?
Imagine you’re planning a vacation. Today, you have to find flights yourself, pick a hotel, check the weather forecast, look up entry requirements, arrange insurance, and put together a packing list. You have to initiate all of that yourself, in every app.
In the future, apps will interact with each other through an AI agent and jointly solve the user’s tasks. For example, you simply tell your voice assistant you’re planning a trip for certain dates, and it takes over from there: analyzing your calendar, finding suitable flights, factoring in your preferences from past trips, checking the weather, suggesting a hotel, handling the necessary paperwork, and reminding you of anything important before departure.
– That’s an impressive prospect, but it immediately raises a question: what if the AI agent makes a mistake, books the wrong hotel or mixes up the dates? Who’s responsible then, the app developer or the user?
It really comes down to the agent’s settings. It might only suggest options and leave the final decision to the user, or it might act almost fully autonomously, right up to booking and paying. That’s why it’s important to define the boundaries of its authority in advance.
An agent’s mistake can come from misinterpreting a request, not having enough context, or hallucinating. For instance, booking a round-trip ticket when the user only wanted one-way. In that case, responsibility is shared: the user is responsible for checking the final details, and the developer is responsible for the quality of the model’s checks and for an interface that clearly explains what the agent is doing.
But there’s a nuance developers are actively discussing right now. Once there are many agents interacting with each other, the chain of actions gets longer. Say one agent books the hotel, a second books the flights, and a third arranges the transfer. If something breaks down along the way, it becomes hard to tell where the fault lies. That’s why Apple and other major players are building logging and transparency into their frameworks. Ideally, the system should be able to show: “I did this, based on this data, and this is where the decision was made.” Then questions of responsibility disappear, because you can always trace the agent’s logic and step in if needed. But for now, that’s more of a future goal than current reality.
– Understood. Let’s come back to the present, then: you’re currently rolling out App Intents at inDrive. What will a typical taxi order through Siri look like?
Right now, the usual flow is: the user opens the app, enters a pickup point and destination, chooses a fare, checks the details, and confirms the ride. In the future, part of that flow could potentially become less visible to the user.Â
For example, someone might just tell Siri, “I want to go home” or “I need to get to the office in 20 minutes.” In that case, the system could infer the context: where the user is, which places they usually go to, what settings they prefer, and pass that information to the app. Since iOS already analyzes anonymized usage patterns, it’s reasonable to imagine that it could eventually become better at suggesting the right app or action at the right moment. Say a person regularly orders an inDrive ride every Thursday after work; at some point, the app itself will start suggesting that same route. All the person has to do is confirm the ready-made action.
If apps evolve in this direction, they will likely become simpler for users. For developers, however, they are likely to become more complex, since a wider range of scenarios will need to be supported.Â
– Speaking of scenarios, you mentioned that inDrive managed to cover around 70% of one module with automated tests, and that’s with the app running in dozens of countries. AI multiplies the number of scenarios by the hundreds. How do you handle that?
In classic mobile development, things are fairly predictable: if the user taps a button, we know the expected outcome and can verify it with automated tests. That’s why high test coverage helps catch bugs before release. With AI, it’s more complicated. The same request can be phrased in different ways, and the model’s response can vary. So it becomes important to test how the system understands user intent, handles ambiguous requests, and behaves in edge cases.
For that, we also rely on automated tests. AI helps generate them, and engineers evaluate the results and make the final calls. This approach is already paying off: we’ve managed to bring UI test stability on iOS up to 94.5%. More broadly, that’s the direction our engineering culture is heading in. Agentic AI at inDrive already handles a significant share of testing work, and one of our current focus areas is training models to run end-to-end mobile tests directly from plain-text instructions.
– How do you approach data security, to protect user privacy in the app?
The first thing to understand is that AI on its own doesn’t automatically mean losing control of data. It all depends on how the work with information is structured and which technologies are used. For developers, this has become one of the key challenges: building services that aren’t just smart, but also secure enough for users to trust.
For instance, when working with AI agents in a service, you can use models hosted locally, meaning they run within the company’s own infrastructure. In that case, data isn’t sent to external servers run by Google, Anthropic, or OpenAI; it stays inside the system.
This approach makes it easier to control user information and to know exactly where it’s being processed. Local models can also be more cost-effective, since data is processed on the company’s own infrastructure rather than requiring constant calls to external APIs.
– Staying on the topic of processes: given the scale of your team, and the fact that colleagues regularly come to you for advice as a senior developer, I’m curious about your view on the future of the profession. With AI agents now helping to write code, do you think developers will gradually turn into managers who coordinate AI, or will the core of the job stay the same?
I don’t think developers will turn into managers, but staying just a programmer who writes code won’t be enough either. The profession has always been evolving, really. Twenty years ago, a big part of the work involved memory management, performance tuning, and other low-level tasks. Today, a lot of that is automated or handled by modern tools. That freed developers up to focus more on architecture, product decisions, and user experience, and now, on working effectively alongside AI as well. AI agents are becoming our helpers, and the key skill now is knowing how to delegate routine work to them while still keeping control over the important decisions.
At the same time, today’s developer is becoming more of a generalist. You need to understand not just programming, but a bit of product development, analytics, design, and user experience too. This is especially true on small teams, where one person often takes on several roles at once. And AI has sped that process up. Today, a specialist can use AI agents as an extra team: setting one agent up for analysis, another for writing code, a third for reviewing solutions. Each one has its own task, instructions, and area of responsibility.
That’s why new skills tied to AI tools keep emerging. The industry is moving very fast: not long ago, everyone was talking about prompt engineering, then more structured approaches took over, like skills and detailed instructions for different scenarios. Now, more and more attention is going toward orchestrating AI agents and managing how they work together. But that phase will probably give way to new approaches too. In AI, trends tend to either become the standard or fade away pretty quickly.
I think the biggest shift is that the developer of the future isn’t just someone who writes code. It’s someone who knows how to define problems, make technical decisions, manage tools, and build complete products.
