What Happens to Digital Products After AGI?

Google DeepMind’s From AGI to ASI paper explores what could happen if artificial intelligence progresses from broadly human-level intelligence towards systems capable of outperforming even the best human experts across a wide range of tasks. While the timeline and eventual capabilities remain uncertain, it raises an important question for anyone building digital products today: what happens to the products we’re creating now as the intelligence underneath them becomes more capable?

You don’t need to predict when AGI will arrive to think about that. Products being designed today could still be operating five or ten years from now, while the models they depend on may change many times during that period. Capabilities that currently require carefully designed workflows, multiple systems and human intervention could eventually become something an AI system can handle largely by itself.

That presents a significant challenge for AI-native applications. How do you build products around technology that is likely to become considerably more capable over their lifetime, without knowing exactly what that technology will be able to do?

Building for AI that keeps improving

DeepMind outlines several possible routes from AGI towards more capable systems, including continued increases in computing power, new architectures, AI systems contributing to the development of subsequent generations and increasingly capable specialised agents working together.

The exact timeline matters less to product teams than the likelihood that the capabilities of the technology underneath an AI product will continue to improve.

Traditional software has generally been designed around relatively fixed capabilities. If you build a calculator into a product, you know what it can do. If you create a workflow, you can define each step and the rules that determine what happens next. The software might be updated, but its fundamental capabilities don’t suddenly change because an external provider releases a new model.

AI is different. A product built around a model today could gain significantly greater reasoning, multimodal or agentic capabilities simply by moving to a newer generation. Tasks that previously required bespoke logic or human intervention could become capabilities of the underlying intelligence itself.

This raises an important product strategy question: how much should you build around what AI can do today, and how much should you account for the possibility that it will be able to do considerably more in the future?

There is a risk in designing too precisely around current limitations. A workflow might contain several stages because the AI currently needs a person to check its work at particular points. If future models can perform those tasks reliably, parts of the original product design may no longer be necessary.

That doesn’t mean removing human oversight in anticipation of more capable AI. There will continue to be decisions where people need to remain responsible, particularly when the consequences of getting something wrong are significant. The important distinction is between human involvement that exists because of a genuine business, regulatory or ethical requirement and involvement that exists because today’s technology has limitations.

Products need to be designed so that these elements can evolve independently. As AI capabilities improve, teams should be able to reconsider workflows and levels of autonomy without having to rebuild the entire product.

Rethinking how people interact with software

The same challenge applies to how people interact with digital products. Most software has traditionally been built around people navigating screens, menus, forms and predefined journeys. Even many AI products still follow this model, with a chatbot or assistant added as another way to interact with the underlying system.

As AI becomes more capable, that balance could change. Instead of telling software exactly what to do at every stage, users may increasingly describe what they want to achieve and allow the system to work out how to get there.

This shifts the focus of product design towards deciding which interactions are genuinely useful, what decisions should remain visible to the user, where approval is required and how people understand what the system has done on their behalf.

Conventional interfaces will still have an important role. Screens, dashboards and structured workflows remain valuable wherever people need to review information, compare options, make decisions or retain control. More capable AI could simply change when those interactions are necessary and how much of the process a user needs to manage themselves.

Products should therefore accommodate different levels of AI capability. A system might initially guide users through a series of steps, then take on more of that work as its capabilities develop, while preserving the visibility and oversight the task requires.

Where will the lasting value sit?

If the underlying models continue to improve, access to intelligence itself may become less of a differentiator. Competing businesses could have access to models with broadly similar capabilities. What those models know about each business, what they can access and what they’re allowed to do could matter far more.

That places greater importance on proprietary data, integrations, workflows and business rules. As we’ve explored before, the value is shifting to the data and the ability to put business-specific information to work can become a significant source of value.

The foundation model doesn’t necessarily need to be proprietary. As discussed in What Should an AI-Native Business Actually Own?, a business can retain the value of its product through the data, processes, integrations and logic that make it specific to its needs, even when the underlying intelligence comes from an external provider.

This also has implications for how products are built. If a product is too tightly coupled to the capabilities or limitations of one model, moving to a more capable alternative could require significant redevelopment. Treating models as components that can be evaluated, replaced or combined gives businesses more flexibility as the technology evolves.

AI-native product development needs to account for this from the outset. That means keeping business logic, data and integrations adaptable, while ensuring that changes to the underlying models don’t compromise the reliability of the wider product.

As AI systems become more capable of making decisions and interacting with multiple systems, defining what they can access, what actions they’re allowed to take and which decisions require approval will also become increasingly important. These are decisions product teams can make now, regardless of when more advanced AI capabilities become available.

Designing for change without predicting the future

There’s a temptation when talking about AGI and superintelligence to imagine that everything we currently build will suddenly become obsolete. But more capable AI won’t automatically remove the constraints that businesses operate within. Existing systems, regulation, customer behaviour, physical infrastructure and organisational change all move at different speeds.

A more capable model doesn’t mean every process can be automated or every product needs to be redesigned. Businesses will still need to consider whether a particular capability solves a genuine problem, whether customers will trust it and whether the costs and risks are justified.

The focus should be on avoiding rigid products built around the exact capabilities of today’s models. Products that can accommodate new models, changing workflows and different levels of autonomy can take advantage of improvements without requiring fundamental redevelopment each time.

This also means making deliberate decisions about what should remain stable. Business rules, permissions, accountability and essential user controls may need to endure even as the technology underneath them changes. Separating these elements from the model itself gives product teams more room to adapt while retaining control over how the product operates.

For businesses investing in AI today, these are practical product considerations rather than predictions about AGI. They influence architectural decisions, product design, the choice of models and how much flexibility to build into a system from the beginning.

Building products that can evolve

You don’t need to build for superintelligence or make product decisions based on a prediction of when AGI might arrive. But you do need to recognise that the AI available when your product launches may be considerably more capable than it was when you started designing it, and that it could change many times during the product’s lifetime.

For businesses thinking about AI now, that means focusing on the problem they’re solving, building around the capabilities available today while allowing for future improvements and retaining control over the elements that make their products distinctive.

Products that can adapt to that progress won’t need to be rebuilt around every new generation of AI. That requires thoughtful decisions about what to build, what to keep flexible and where human judgement remains essential.

The aim isn’t to build a product for AGI. It’s to avoid building one that assumes today’s AI is as good as AI is going to get.