January 15, 2026
In his seminal 2010 article on responsive designEthan Marcotte described a new way of thinking about web user interfaces that went beyond a fixed frame. In response to his customers’ requests for an “iPhone version” of their websites, he instead proposed a single experience that could answers to the user’s device. Page elements can use a different set of design properties for large and small screens, transforming their shape to match the capabilities of the hardware. Crucially, on simpler devices, elements can be completely hidden if they are deemed non-essential.
As this technique became popular among designers and product teams, they quickly realized that starting with the “full” version and slimming it down was more challenging. Instead, you can start with the least With a capable device, you could strip the product offering down to its bare essentials and present those primitives on the limited screen real estate available. This quickly led to recognition of the irony of responsive design: that the least capable devices often had the most effective user experience because they were most closely tailored to the user’s needs.
We came to realize that responsive design wasn’t just about layouts, it was about forcing organizations to confront what really matters. But organizations, it turns out, can’t resist the temptation to use available UI space for promotions or displays of their org chart, all of which almost always get in the way of what users are trying to do. That this is still the case all these years later is astonishing; the difference in the UX of apps versus websites for banking, airlines and e-commerce is clear evidence.
Towards a more radical responsiveness
I believe we are at the dawn of a new era of further crystallization, and it feels both more disruptive and more enabling than the shift to responsive design 15 years ago. Earlier last year, basic model providers released coding agents that developers could install and use directly from their command-line terminals. These agents, including Claude Code and OpenAI Codex, introduced two seemingly minor improvements that would prove to be incredibly important.
First, they could repeat. Unlike previous chatbot interfaces, these agents continued to work and refine their output, moving through a plan until goals were achieved. Second, they gained access to resources. They could now call standard Unix utilities such as curl for loading web pages or grep for searching file contents. A few months later, the underlying models were updated specifically to take advantage of this type of tool usage and stay more focused on plan tracking. With these pieces in place, developers at the frontier of AI-assisted engineering started sharing new workflows that felt truly different, and things started to pick up speed again.
Recently, many non-developers, including myself, have discovered that using Claude Code with files locally can be an incredibly effective way to get work done. My social feed is full of people sharing their use cases: putting the agent to work in an Obsidian vault, managing email and calendars, and finally getting value from smart home devices.
Primitives all the way down
The key to all this is the tools: small, simple command-line apps with clear, direct documentation. Despite the promise of MCP servers, Claude Code has access to iCalBuddy is a much faster and more effective way to ask for help organizing next week’s schedule. Codex is an expert in the field ghthe CLI app for GitHub, and can quickly summarize and organize your next sprint’s open issues. Both can quickly navigate 1Password with the op command to prevent access tokens from being distributed everywhere. These apps clearly expose the primitives of the systems behind them. An agent that iteratively creates a loop while connecting dozens of these composable tools together is starting to feel like a whole new way of working.
This is the next step in responsive design and the blueprints are lurking all around us. The coding agents now let you add skills: simple descriptions of how to perform a task, written in natural language. For example, you can explain to your agent how to pull data from an internal API to help you create your quarterly update. But if you really want to get a good idea of what the future holds, try this: On an iPhone or Mac, open the Shortcuts app and start searching through the available commands displayed by each app you have installed. There you’ll find a remarkably clear visualization of all the atomic components of an app’s capabilities. All nouns and verbs.
None of them That doesn’t work with agents yetbut I think this is the clearest glimpse into what apps will become in a world of advanced agentic workflows. Here, in three screens, are the actions we can take when we communicate with friends, plan our vacations and manage our money. This is what it looks like when apps are honest about what they can do. And one step further: all the APIs for all business SAAS and personal productivity apps, lurking behind developer documentation, waiting to be unlocked. What is a “seat” and what is a “user” when we send our agents to ask questions and make requests? This isn’t just app design, but what companies become as we collectively lean harder on agents.
Clarity as a competitive advantage
This may seem counterintuitive, but my career in user experience design makes me even more excited about a potential future like this. If agentic workflows strip applications down to their essence, then it makes sense that they present what they find in UI components created on the fly. Perfect customization, whether you access the capabilities on a big screen or through your AirPods, in English or Cantonese, wrapped in shadcn or Chakra UI: your choice!
An agentic future elevates design to pure strategy, which is what the best designers have always wanted. Creating a great user experience is impossible if the way the company expresses its capabilities is unclear, vague, or misleading. So the best have always pushed for more authority and more access to the conversations where the real decisions are made in their organizations. Gaining that access while preserving the interests of people who use the products has always been my goal as a designer.
This requires quite a leap of faith for those who practice design. I see the same thing happening with engineers who are waking up to the fact that they will soon stop typing code, and not long after, stop reading it. Many of us imagine a future where aspects of the work we love are quickly slipping away. But I will argue that this is what we have always been: responsive designs that mold to our customers, patients and citizens in a way that asks companies and institutions to express exactly what they have to offer.
If an agent were to use your product tomorrow, what truths would they discover about your organization?
#coding #tools #future #design