Stop adjusting settings. Just tell it how hard to think.

On January 9, 2007, the first iPhone was introduced with a screen that responded to movements people already knew. You tapped something. You flicked a list. You pinched a photo. And the gestures weren’t simply decorations laid over the controls. They were the controls.

The launch described an interface that you ran with your fingers instead of a stylus or a keyboard, and the idea underneath it mattered far more than the hardware did. People no longer had to translate a simple intention into whatever sequence the device demanded. Instead, the device met them in a language their hands already spoke.

That one principle shaped the next two decades of consumer technology, because it stripped out a whole layer of translation between what you wanted and what you had to do to get to it. The strongest interface may be the one that keeps controls out of the way and lets people express what they want to accomplish as naturally as possible.

And yet LLM-based AI has brought some of those settings-style choices back into the experience. This time, the question is less about something like zoom level and more about how much reasoning a particular task may require.

AI brought the settings panel back

A quick fact and a life-changing decision both get typed into the very same box. But they don’t deserve the same amount of thought. One just needs a fast, correct answer. The other may call for a broader view, additional cross-checking and a willingness to revisit the initial conclusion.

Some of today’s LLM AI now offers reasoning modes, token budgets, or effort controls stamped low, medium, and high. And that’s genuine progress, because it reflects something true about the technology. More computation during inference really can produce a better answer, especially when the method and the effort are matched to the difficulty of the prompt.

A widely discussed study of test-time computation found out that allocation strategy mattered. Different approaches behaved differently as the difficulty changed, and spending compute adaptively, where it was actually needed, worked better than simply throwing more of it at every problem.

The consumer takeaway may be simpler than the underlying research. More effort can be useful, but asking people to choose a technical mode before they fully understand the problem can add unnecessary friction. It is a bit like asking a first-time skier to choose a run before they have a clear view of the mountain.

Longer isn’t the same as deeper

You’ve probably already typed think hard, or be thorough, or don’t rush. And conventional LLM AI tends to react to that. The answer gets longer, it adds a few steps, and it takes on a more serious tone. But none of that proves that the actual thinking underneath changed in any real proportion to the stakes.

Sometimes the extra words simply restate the same premise, making an answer appear more considered without necessarily reflecting deeper analysis. Meanwhile, the actual level of effort may still be determined by a mode, token allowance or other setting that many users never encounter.

And that leaves the user holding an odd kind of burden. You have to phrase the question, guess at how technically hard it is, understand the product’s controls, and then decide how much compute to ask for. People rarely need to think about the technical mechanics behind intuitive digital interactions, and requesting careful or even basic reasoning from an AI system could work much the same way.

The stakes are already in the sentence

People signal effort all the time, and usually without even thinking about it. “Give me the quick version” means you care about speed more than you care about getting every last detail. But something like, “before I wipe my old phone check that these photos are actually backed up”, means the cost of a careless answer just went way up. And “look at every failure point before I do this” tells a co-worker to widen the search and challenge every obvious route.

Nobody needs a menu to tell those three apart. The urgency, the depth you want, the room for error, all of it rides along inside the request you’re making. Human language hands over the task as well as the level of attention the task deserves in the very same breath.

And that’s the thing a natural interface for intelligence ought to protect. The sentence shouldn’t just change the style of the reply. It should change the amount and the shape of the thinking behind it.

Effort should change the structure

Vertus treats effort as part of the meaning. Inside its Cognitive Reasoning Superintelligence, Effort-Driven Cognition is designed to use ordinary language as a signal for how much processing a task may require. Rather than selecting a separate technical mode, users can describe what they need in familiar terms, with the phrasing helping shape how the system approaches the request.

The mechanism behind it is called Effort-Responsive Processing. A request for a harder look, for example, may lead Vertus to use a broader processing structure, with more integration and additional internal stages, while a request for speed may favor a more streamlined path. The idea is that the difference would appear in how the work is carried out behind the response, rather than simply as a longer answer on the page.

A faster request can call for a narrower level of analysis, while a higher-stakes question or a request to test competing conclusions may call for a broader approach. Vertus is designed to adjust to those cues without requiring the user to translate them into technical parameters.

This does not depend on reading a user’s mind. The relevant cues are already present in the way a request is phrased, and Vertus is designed to treat differences in emphasis and intensity as signals rather than smoothing them into the same response pattern.

The best control is the one that disappears

The first iPhone didn’t forever get rid of settings. It just lifted the common intentions out of them. Tap, flick, and pinch became enough for the things people did most, and all the machinery slipped back behind the experience.

Cognitive effort could benefit from the same kind of responsiveness. A request to “think deeply” might reasonably lead to broader analysis, while “keep it simple” could signal a more focused approach rather than merely a shorter version of the same process.

Intelligence could be easier to use when the system responds to the task and its stakes without requiring users to manage the underlying mechanics. The level of effort might be guided by natural language rather than a separate set of controls.

Digital Trends partners with external contributors. All contributor content is reviewed by the Digital Trends editorial staff.