For pretty much as long as software has existed, using it has been the price of getting something done. Want to pay a bill? Open the banking app. Want dinner? Scroll the delivery app. Want to book a flight? Click through departures, seat maps, fare classes and upsells for travel insurance. Engagement measured the success of a product, but it was also necessary to complete a task — people were going to engage with products no matter what. Products didn’t so much have to create engagement as they had to capture the highest share of it.
Agents change that dynamic. For the first time in software history we can get the outcome without the interaction; the bill gets paid, dinner gets ordered, the flight gets booked, and we never have to open the product (if there even is one) that facilitates the task. The choice to engage with a product or not simply didn’t exist before, and I think it significantly changes how we value engagement. Products move from value capture to value creation. Take away the necessity to use a product and you find out how much of that engagement was ever really owned.
Theoretically, when a handful of LLM super-apps/assistants own nearly all engagement across every digital facet of a customer’s life, such as shopping, banking, transportation and other day-to-day tasks, businesses built as the interface layer on top of infrastructure become obsolete, and businesses that do own infrastructure simply become dumb pipes facilitating transactions on a centralized interface or an agent. The engagement and relationships that products have with customers are at risk, and that forces us to rethink what we’re designing for.
The best interfaces have a dual mandate: they need to be both efficient and usable, but also expressive and experiential, giving users a perspective on the problem they solve. They are about feeling as much as they are about doing, and that’s what differentiates a tool that gets a task done from an experience that does so in a way that engages the human.
Super apps & super stores
Imagine telling someone that for the rest of their life they could only shop on Amazon. Other stores would still exist; they just couldn’t go to them. No farmers market, no luxury retailer, no bodega. Most people would feel that as a real loss, despite the fact that Amazon can get them almost anything by tomorrow and at a great price.
When e-commerce marketplaces showed up and big box retailers went digital, there was a time when we figured small independent stores were no longer viable. I mean, how could smaller retailers compete on things like price and convenience? As we now know, indie stores didn’t just not die — many thrived and some scaled up. The threat of large retailers forced them to zero in on what their strengths were and where they could win customers. What we learned was that customers aren’t one-dimensional. Yes, they like speed and convenience, but they also deeply value character, familiarity and perspective. An indie retailer’s perspective on their offering is felt through everything: store design, location, staff, assortment, merchandising, pricing, etc. All of it communicates to customers not just how a store solves a problem, but how and why it thinks that problem should be solved. That communication is what creates a durable connection with customers.
While retailers like Amazon made going to a store optional, the stores that thrived knew they couldn’t just exist; they needed to deliver an experience that was worth going to. We probably visit our local boutique a lot less often than we order from Amazon, but when we do go, we go on purpose. Both kinds of stores are recipients of consumer spending, but only one is the recipient of customers’ attention.
I think it’s likely we end up in a similar place in software, where agents play the same role that Amazon plays for shopping; they become the de facto destination for everything we don’t really care that much about. I’ll happily let an agent fill out forms on my behalf, the same way I’ll buy paper towels on Amazon. But when it comes to buying an important gift, or something that’s maybe difficult to return once I buy it, Amazon doesn’t (usually) work.
Where assistants create value
The Muse and Instincts of the world offer enough visibility into their interactions, and engender enough trust to handle low-consideration tasks, or things that can be revised later. These interactions are without a doubt useful, but they don’t represent a particularly valuable slice of what humans spend their time thinking about. These assistants’ real value is not so much in taking meaningful interactions off our plates, but in making cognitive space for us to spend more time with the things we do care about. We might have always cared a lot about cooking healthy meals, but didn’t have enough time to comb through a cookbook because household admin tasks kept us too busy or too distracted. It’s easy to think that an agent will help on both fronts, taking care of the admin tasks and distilling cookbooks for us — and it can — but I think what we’ll find is that humans will choose to spend time reading the cookbook.
One of the flaws in the argument for the collapse or consolidation of interfaces into super-apps or purely generative interfaces is that it assumes there is a single correct way for software to be built and used. It assumes that because an assistant can handle any task, people will want to give it every task, or that because generating hyper-personal UI on the fly is possible, it will always offer the best interaction model. We make these assumptions because, up until this point, software has largely been limited to a single mode of interaction that required the total involvement of the human. When that’s no longer true, we need to adapt our products to a world in which ‘using software’ means anything from handing an agent the whole task to handing it just parts of it. Maybe a user wants the agent to book concert tickets, but wants to spend time picking the seats, and then passes the task back to the agent to complete the purchase. Users having a choice in software doesn’t just mean choosing whether to use it, but also choosing which pieces of a task they want to engage with. As software users, we haven’t yet been challenged with thinking about what software we do and don’t care about using. But as more and more software is able to ‘just work’ without much engagement, we’ll have the opportunity and the responsibility to become more judicious and selective about which products and tasks get our attention and engagement.
How this changes design
So what does this mean for product design? How does choice impact how we design products that users want to spend time with? I think there are three core changes that we need to be aware of.
First, if people are choosing to spend their time and attention on a product, that product needs to reward them in some way. A product worth paying attention to should leave you more interested/interesting than it found you.
That doesn’t have to be about intelligence. A specialty boutique might stock something you genuinely can’t find anywhere else, and that item might become part of your personal style. Knowing it exists, understanding why you like it and what it means to you — that’s a kind of knowledge, and it makes you a little more you.
What makes this possible is context. We’ve spent the last few years figuring out how to give agents the context they need to do good work: memory, data, connectors to all of our tools. But humans need context too, and we get it not just from basic feedback, but from the entirety of an experience, and delegation to an agent assistant strips that away. When an agent completes a task for you, you get a compressed result, not the texture behind it. You get the item, but not the whole story; what the alternatives were or why it was the right one.
Deep, specific product experiences are where that context lives. The experience delivers context, context builds perspective, and perspective is what turns a purchase into part of your style, or a task into something you actually understand. And all of this actually makes us better at utilizing agents. Someone with richer context gives sharper instructions and is better at judging the results. The products worth choosing don’t compete with agents, they make us better at directing them.
Second, we will very likely see fewer interactions with our products. I briefly mentioned this earlier, but we need to be prepared for a world in which a user might engage with the product on an irregular cadence or only in pieces. Every facet of the product needs to not only stand on its own, but deliver outsized value and experience without relying on the whole. We’ll need to design for fewer but richer experiences for a (hopefully) more focused user.
Just like our Amazon example; for most people, going to Amazon or a big box retailer is convenient for everyday purchases, but convenience and standardization come at the expense of experience. We might go to our local boutiques and specialty stores less often, but we make a point to go because they provide a differentiated and unique experience. Even if these stores don’t have the scale that large retailers like Walmart or Amazon can offer, the touch points they do have with customers are meaningful and highly valuable.
Third, we’ll need to design the human back into our products. As assistants and agentic software become more common, we’ll start to notice a perceptible line between interacting through an agent and interacting directly with a product interface. Intentionally using software will feel different. If a user wants something done effortlessly, interacting with an agent delivers beautifully on that use case — and similar to how assistants taking our low-value tasks away creates more cognitive space, I think the existence of agentic software will actually give designers more freedom and bandwidth to focus on creating more human, textured experiences that prioritize feeling and impact over the scalability and systematization that have dominated this most recent era of industrialized software.
For decades now, we’ve designed products to abstract humans away in favor of scalable systems. Standardize the flow, remove the friction, automate the decision. That playbook also happens to be exactly what agents do better than any interface ever could. A product that try to compete on efficiency is now competing with agents, and that’s not winnable.
The gap between machine-made and human-made isn’t really a technological one; a machine can (or will) absolutely create something that looks indistinguishable from a human creation. The real difference is how we interpret it. Do we understand what a product is saying, why it exists or how it behaves? Does it feel like a human is on the other side of the screen? That isn’t always critical for successful software, but for software that wants to built and grow a relationship with a user, it absolutely is.
Invest in your users
The pushback on all of this is: when (not if) engagement becomes optional, there won’t necessarily be a healthy split between using agents and using software. It’s entirely possible that agent and assistant use represents 80% of all software use — and does that leave too little engagement to sustain most products? I think that’s certainly a risk. At the same time, scarcity raises value. Designing with the goal of creating more of a scarce resource adds a valuable dimension to design as a practice.
A single intentional, chosen interaction with a product says more about loyalty and value than a hundred free or captive ones. The more software disappears into the background, the more the software that we choose to look at needs to mean something; as agents make more space in our brains, we’ll fill that space with things we care about, and how we define a ‘meaningful’ task will likely evolve.
Investing in, developing and owning the relationship with users, is more critical than ever; and the products that leave us with richer context and a sharper perspective are the ones we’ll keep coming back to.

