At 8:15 on a weekday morning, a personal AI agent should be useful even if its owner is holding a coffee, searching for the right train and wondering whether the 9 a.m. meeting has moved. Yet much of today’s agent race is being fought inside chat windows, dashboards and messaging threads—places that still ask people to stop what they are doing and look down.
The progress is real. Unlike chatbots that mainly respond with information, personal agents can connect to services, carry out multi-step tasks and return with results. But asking what an agent can do only answers half the question. The other half is how comfortably it fits into an ordinary day.
Three routes into the personal agent market
Meta’s Muse is the broad-platform approach. Built to work across the apps people already use, it handles tasks ranging from scheduling to shopping and travel bookings. Its dedicated app and WhatsApp integration give it familiar points of entry; a cloud-based virtual computer lets it operate beyond a single service. Muse also reflects a crucial design choice: consequential actions, such as purchases, need user approval. The advantage is scope. The challenge is making that breadth feel simple rather than like another layer of software to manage.
Instinct takes a different route. It has no standalone app. People interact with it through texts, calls or WhatsApp, while the agent works across digital services on their behalf. Travel has emerged as a particularly strong use case. In September, TechCrunch reported the founder’s statement that more than half of the platform’s transactions were travel-related. Instinct is not a travel-only agent, but that concentration is telling: complicated bookings make the value of delegation easy to grasp. Its phone-call option also shows that the move beyond a conventional interface is already under way.
Town offers a third model, built around existing workflows. It helps with email, calendars, documents and the small follow-ups that fill a working day. Although Town increasingly presents itself as a work assistant, its lesson applies beyond the office: much of what people want to delegate is not glamorous. It is finding a meeting time, drafting a reply or remembering to follow up. The agent becomes valuable by removing dozens of small decisions rather than one enormous task.
The boundaries overlap: Muse acts proactively, Instinct accepts calls and Town extends beyond the inbox. What distinguishes them is where each puts its weight: broad capability, task execution or workflow integration.
The next interface must work when life gets in the way
For decades, software has assumed we can give it our eyes and hands. That assumption becomes less useful while cooking, walking, commuting or getting children ready for school. A graphical user interface (GUI) demands visual attention. A language-based interface (LUI) makes instructions easier to express, but when that language lives in a text box or chat thread, the user still has to read, type and check the response.
Voice offers a way out, provided it is treated as more than an input method. The opportunity is not merely dictating prompts we would otherwise type. It is building an agent that can start a useful conversation, accommodate interruptions and help someone move from information to action without shifting to a screen.
Orilo is an early example of this voice-first approach. It starts with personalised audio briefings shaped around a person’s interests and daily context, rather than a list of commands waiting to be entered. A listener can ask why a story matters, follow a tangent or move on. With appropriate permissions, Orilo can bring calendar information into the conversation and help add reminders or events when something worth acting on comes up.
Consider a morning commute. A conventional agent might wait for “Summarise today’s news,” deliver an answer and leave the user to compose the next request. Orilo’s model is more conversational: offer a relevant starting point, take a follow-up question and capture a reminder before the journey ends. That may seem like a modest change, but it makes otherwise occupied time useful without turning it into screen time.
Voice-first is more demanding than voice-enabled
Building that experience requires more than adding speech recognition to a chatbot. Spoken interaction exposes latency, awkward turn-taking and weak memory immediately. An agent must know when to speak, when to stop, which context is relevant and when a request should trigger a tool rather than another paragraph of explanation.
Trust matters just as much. A daily briefing can be offered proactively; sending a message, changing a meeting or spending money calls for explicit authority. Users need a way to inspect important actions and correct mistakes. And some tasks—comparing flight options or reviewing a contract—are simply easier on a screen. Voice-first should not mean voice-only.
There is room for all these approaches. Muse demonstrates the appeal of an agent that can do almost anything. Instinct shows the value of solving concrete problems without another app. Town focuses on repetitive work in familiar tools. Orilo is testing a different proposition: that the most personal interface may be the one people barely have to look at.
The more meaningful measure of success may not be how many tasks an agent can complete in a demo. It may be how often it helps someone get something done without having to stop living their day.
