Part 3 of 3 in our series AI Agents in Design.

Most technology is designed to hold our attention. Carlo Edoardo Ferraris, CEO of Natura Umana and co-founder of Rolling Square, is building AI that does the opposite: it works quietly in the background, speaks instead of showing a screen, and nudges people back towards each other. Natura Umana, the AI lab created by Rolling Square, makes HumanPods, earbuds that give voice access to a team of AI agents called “AI People” through its NatureOS app. Carlo explains the thinking behind it, how they measure success, and the risks designers need to face.

“NaturaUmana’s agents do not trap users in virtual ecosystems; instead, they promote interactions with other humans when beneficial.”

Natura Umana’s mission centres on “rediscovering human nature” through technology. How does that shape your approach to designing AI agents, and which ethical frameworks guide you?

Natura Umana’s mission to rediscover human nature means unlearning the habits we’ve developed through excessive technology use, and adapting tech to fit natural human behaviour rather than forcing humans to adapt to it. This shapes how we design AI agents: we focus on voice-based interactions that mimic natural human conversation, minimise distractions, reduce screen time and combat the addiction caused by digital consumption.

To understand this approach, it helps to recognise what needs to be unlearned. Our interaction with personal technology began as functional and convenient: phone calls, emails, browsing, learning and shopping. But as neuromarketing strategies captured our attention, excess emerged. Today we spend roughly 40% of our waking time in front of screens, 60% of it on entertainment. The virtual world, created for convenience, now often drains energy without giving anything back. This isn’t inherent to human nature: convenience has led us to step back from natural interactions with other people, with nature and with ourselves.

AI has the power, resources and reasoning to work for us, solving problems while giving back the time we hand over to machines. Two conditions guide AI agent design at Natura Umana:

The AI agent acts as an extension of the user, working efficiently in the background without demanding the user’s time unnecessarily. This needs short-term and long-term memory, integrated into the user’s ecosystem.

The AI acknowledges human tendencies and guides users towards more natural behaviour, such as encouraging them to do things without technology when that’s appropriate.

Our “AI People” are engineered to emulate human behaviour, with unique personalities, voices, knowledge, memories and beliefs, so interactions feel natural and evolve with the user. For instance, the primary AI Person, Nature, assesses each query and connects the user to specialised AI People, such as therapists or coaches, giving personalised, human-like support without overwhelming them.

Our ethical frameworks put responsible development first: improving human life as a whole, even if that sometimes means less technology. That includes non-invasive augmentation, ethical innovation in natural language processing and intent understanding to break attention-harvesting cycles, and making sure technology enhances life without harm. We respect human boundaries by focusing on balanced, focused lifestyles and transparent practices that avoid dependency.

Given your goal to reverse addictive technology, which design principles and methods help AI agents provide value without creating dependency? How do you measure success?

Our design principles create a synergy between hardware and software. A core principle is offering AI through wearables rather than smartphone screens: ambient, non-invasive interaction, with voice as the main interface, that keeps people present and aware of their surroundings. For wearables to really combat addiction, they must be comfortable, non-invasive to human interaction, and give true screen-less access to AI.

HumanPods, our first wearable, are built on this:

Comfortable: a gravity hook and ultralight design let them rest just outside the ear canal, so you easily forget you’re wearing them.

Non-invasive: the ear canal stays open, so you stay engaged with real-life surroundings.

Truly screen-less: with NatureOS, the phone can stay locked in a pocket; a double tap on the Pods activates the AI for fully screen-less tasks.

Software principles complement this. First is background reasoning and execution, so the AI handles simple tasks without screen checks or a constant voice connection. Another is using multiple agents in a collaborative system: one main AI Person, such as Nature, orchestrates specialised AI People and delegates tasks to them, completing them in the background without the user overseeing every step. Complex workflows are broken down and handled autonomously, which makes the system proactive and further reduces dependency. Full integration with tools like Gmail and Google Calendar is also foundational, so agents can schedule events, send emails or pull information without pulling the user back into screens or apps.

The interfaces are radically simple and familiar, such as a chat-like view with automatic transcription, with specialised models for specific roles. AI People have memory and learn, which cuts repetitive input, and the hardware integration gives a “wear and forget” experience. The focus is on enhancement rather than compulsion.

For us, success means users spending less time on screens and living in a happier, more natural way. In the long term, we would like to see birth rates rise and depression fall. For the system itself, we track technical measures such as response latency of 1,600 ms, a word error rate under 10%, and a 1.4-second start time, for natural, frictionless interactions.

Your vision treats voice as the most natural way to interact with technology. What unique considerations and challenges come with designing agents that talk rather than show?

We design our agents for near-instant responses with high accuracy, so conversations feel effortless and human. Key considerations include giving agents distinct voices, personalities and conversational nuances, so they don’t feel monotonous and suit a wide range of users, and memory that lets dialogue carry context and evolve over time. The hardware adds environmental awareness, such as noise cancellation of up to 90% using environmental noise cancellation and dual-mic algorithms, and automatic bass adjustment for clarity at low volumes.

The challenges include:

• keeping response times fast enough for a natural flow;
• building a whole software system that works without constant direct input, under the guidance of the main AI Person following the user’s requests;
• creating distinctive agents for diverse user needs;
• teaching the AI to recognise moments when real-life human interaction might happen, so it pauses and waits for the user’s attention to return without interrupting;
• routing each query to the right agent with no visual cues;
• the occasional need for the phone for visual input, such as a photo for context, which works against our goal of fewer screens.

Tap gestures help in voice-only situations, while fitting different ear sizes and staying comfortable all day are hardware challenges of their own.

Natura Umana combines hardware (HumanPods) and software (NatureOS). How does that shape your design process, and how do you keep the experience seamless?

The integrated approach creates one ecosystem: HumanPods give ambient, voice-driven access to AI People through NatureOS, so agents can be always on without interrupting. The agents use the hardware for real-time awareness of the surroundings and the software for memory-based, personal responses, adapting to the user in a “wear and forget” way. A key idea is having one main AI Person that manages the other agents, which is crucial for offering a wide range of capabilities through voice alone.

The connection between HumanPods and the NatureOS app for iOS and Android provides low-latency voice processing, automatic transcription and support for text, images and voice notes. When the app is needed, voice and screen must work together, for example by giving a transcript or other details on request: the AI writes the requested information into the chat. Advanced memory, which works across both voice and screen, is what we see as setting Natura Umana apart. Partnerships with AI infrastructure providers improve performance, and features like quick charging (15 minutes for 9 hours) and gesture controls add to a cohesive, distraction-free experience.

How will the role of AI companions in experience design evolve over the next decade, which capabilities will matter most, and what risks should designers address now?

We see AI companions becoming deeply personal and human-like, with each user potentially having their own adaptive “Nature”: billions of individual companions enhancing daily life through voice and memory. Over the next decade they will shift from transactional tools to intelligent companions that remember conversations, learn preferences and connect users to specialised networks, helping people reclaim their focus and use screens less. This will probably happen niche by niche, in professional and personal use, with AI gathering more targeted knowledge that people can reach naturally. Hardware like HumanPods will matter a great deal, giving hands-free access to AI’s personal knowledge during work or daily life.

The most transformative capabilities will be very low-latency voice responses, personalities that evolve with beliefs and memories, and seamless handoff to specialised AI for things like therapy or coaching. Another is AI that can operate across all our technology, reducing what humans have to do to essentials like voice or touch. Deep integration between platforms and apps can automate tasks and free people for inherently human activities.

“As AI becomes more integrated, ensuring it doesn’t replace real human interactions but enhances them will be critical.”

The risks designers must address include over-reliance on AI for emotional support, for example as a therapist; privacy concerns around memory; and keeping human connections strong even as screen time falls. Designers should handle these through transparent data practices, user controls and clear boundaries that prevent isolation.

How might Natura Umana’s approach influence broader design practice in the industry?

It could encourage voice-first, non-addictive AI design that adapts to humans, and a shift from screen-heavy interfaces to ambient, conversational systems. It may inspire more ethical hardware-software integration, with a focus on low latency, personal AI personalities and memory that make technology less distracting. A major influence will be comfortable wearables that blend into daily life without intruding.

The industry may also embrace multiple agents working together, sharing memories to give individual experiences while handling complex tasks on their own. More broadly, it could redefine AI hardware around sustainability and comfort, break attention-harvesting cycles, and lead to industry standards for reducing dependency. Features like our “Agency”, where AI agents run complete online searches, bookings and solutions with a personal touch, could set a precedent for proactive, user-centred design.

Previously in the series: Kumar Gaurav on designing for intent and Patrick Henz on ethics and governance .