Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- The demise of the classic UI is imminent.
- Salesforce, a bellwether, goes direct to agents with no browser UI.
- With AI, viewable UIs can be delivered “just in time” to users
Recently, Salesforce announced “Headless 360,” in which Salesforce, Agentforce, and Slack platforms are now exposed as APIs, MCP, and CLI to agents, which can access data, workflows, and tasks directly, with no browser user interface (UI) required.
Also: I built two apps with just my voice and a mouse – are IDEs already obsolete?
Salesforce is the bellwether, of course. The future of UI is increasingly geared toward catering to agents, which doesn’t require compelling graphics, clickable buttons, or entry points. This transition was explored by Michael Grinich, founder of WorkOS, who offered observations and predictions at the TypeScript AI Demo Day in San Francisco in April, stating, “We are exiting the UI era.”
Disposable interfaces generated on demand
UIs are evolving from the fixed, static screens we’ve viewed for decades to generated “just-in-time” projection layers that appear as simple text boxes, Grinich stated. In many cases, people will no longer be interacting directly with UIs — applications will deliver results via APIs tied to AI outputs or agents. Interfaces that users see, he explained, will be “disposable — a one-time use interface that just gets generated on demand and then poof, it’s gone. And when you need a new one, just make a new interface.”
This opens a new phase of software development — today’s and tomorrow’s solutions are becoming more self-driven and autonomous. “Software is shifting from these interfaces that you operate to systems that produce outcomes,” he said. “The user expresses an intent, a suggestion, an idea, and from that you send it to the model, and the model is what creates the UI and actions.”
Also: How the rise of AI-native software could give SMBs enterprise-level power
In the process, AI is rearranging the human-computer interface — and, ironically, makes computing more human-centric. Generative AI, one of the fastest-growing technologies of all time, presents a simple text box that asks, “What do you want?” he explained.
UIs have progressed “from switches to commands to pointers, cursors to touch, and now to language,” he said. “Due to language models, we’ve had this breakthrough. where the UIs are now synthesized. They’re generated per request, just in time for you. They’re context aware. They have the immediate information of what you’re trying to accomplish and the world around you.”
4 ways to prepare for the transition
This means a change from the user perspective as well. “The user role has changed here from the operator,” Grinich pointed out — going from simply being a user to that of collaborator and ultimately a director of AI agents.
Grinich provided four pieces of advice to technology professionals on making this transition:
Also: The new rules for AI-assisted code in the Linux kernel: What every dev needs to know
- UI is no longer the product. The product is the capability, model, and data brought together. “The UI is actually just a projection layer of all that. It’s just a way to represent this output,” Grinich said.
- The components still matter. The UI is “not hand assembled anymore; it’s not lovingly handcrafted by people,” he explained. “You’re giving elements to the model, and the model is figuring out what to do with it. It’s a very different interaction paradigm for building UI, because you don’t really know what will be shown to the user. You just have to provide the right type of elements to the LLM [large language model] in the right context for it to make decisions.”
- APIs become the real surface that you’re building on. “The UI is no longer a product — it’s the API,” Grinich said. “Agents don’t really click buttons; they prefer an API.”
- The model is the interface. The interface “is reduced to an API, to a data layer,” said Grinich. “The idea is reducing and reducing and reducing, and trying to make things simpler for people, so there’s less cognitive overload.” Grinich compares this to the ongoing evolution of cars, which have minimized buttons and switches on the dashboard in favor of digital controls and, ultimately, are more autonomous. “You don’t really care about driving. You care about getting to your destination.”
Y Combinator, the Silicon Valley-based business incubator, offers clients a classic single-line instruction: “make something people want,” Grinich related. “I might make a little edit to it: ‘make something that agents want.’ The agents will be doing things for people. If you want to serve people, you need to serve their agents, too.”
