Scout, Remy, Hatch: To Do Everything, the Agent Must See Everything

On June 2, Microsoft unveiled Scout, which it calls its first "Autopilot" agent: software that works in the background, under its own identity, and acts on your behalf without needing to be told twice. The same week brought word that Google was preparing Remy, an assistant tucked inside the Gemini app, and that Meta was testing Hatch, designed to live inside Instagram, where more than two billion people spend part of their day.

Three names, one ambition: a permanent assistant that no longer merely answers, but watches, anticipates and acts. The promise is plain, almost soothing first thing in the morning: hand off the administrative half of life, the dozens of small decisions that nibble at your attention before the first coffee. What remains to be weighed is the price of letting a program hold your days for you.

From the Tool That Answers to the Agent That Acts First

The idea did not come out of a corporate lab. In January, an Austrian developer named Peter Steinberger released OpenClaw, a free tool that let you send a message on WhatsApp or Telegram and have the software do the rest: book a meeting, draft an email, place an order online. Within weeks, 3.2 million people had adopted it, one of the fastest software launches in history. By February, OpenAI had hired its author.

What that wave exposed comes down to a single distinction. A classic assistant waits for a question and answers it. An agent chains several steps together, drives other software and carries a task to completion with less supervision. OpenClaw still had to be installed on your computer. Scout, Remy and Hatch do the opposite: they slip into the apps already open in front of you, the calendar, the inbox, the social feed, where digital life actually unfolds.

The Time an Agent That Anticipates Gives Back

Scout's listed capabilities map out exactly what shifts. The agent schedules and coordinates meetings across time zones, flags the appointments that matter and prepares the relevant materials in advance, spots a looming deadline and blocks out time on its own to meet it, and surfaces a decision left hanging. Microsoft describes a working memory, called Work IQ, that gradually learns your priorities.

The gain is not really measured in minutes. What such an agent claims to return is mental load: the invisible list of things not to forget, the cost of keeping a corner of your mind permanently busy with logistics. For anyone juggling a crowded calendar and an overflowing inbox, the appeal is anything but abstract. Delegating not just the doing of a task, but the very act of thinking about it, changes the texture of a day. It is no longer time clawed back from a chore, but a whole strand of vigilance you agree to set down.

To Anticipate, It Must First See Everything

This anticipation rests on a condition that rarely makes the pitch. To guess what matters, the agent has to watch what does not. Scout reaches into Teams chats, Outlook email, OneDrive and SharePoint files, the calendar, the contacts, and extends to the browser and the machine's local resources. Remy draws on Google's search, mail and calendar. Hatch settles into the heart of Instagram. An assistant that gets ahead of your needs is, by design, an assistant that watches you without pause.

That is the exact underside of convenience. The more the agent knows, the better it serves; the better it serves, the more you hand over. Dependence sets in smoothly, through small successive delegations, until the day you could no longer pick up the thread of a routine you have entirely handed to the machine. Comfort is paid for with privacy thrown wide open, and with a memory of yourself, held outside yourself, that you do not always control.

The Safety Net, Its Cost and Its Failures

Then there is the matter of cost, in the literal sense. Meta is reportedly considering charging up to 200 dollars a month for Hatch. The order of magnitude is not arbitrary: running these agents continuously consumes enormous amounts of compute. Steinberger himself recounted watching a hundred of his agents ring up 1.3 million dollars in usage over thirty days. The assistant tidying your days runs, behind the scenes, on a bill someone eventually pays.

Finally there is reliability, which has become the real question of the year: how long can an agent work on its own before it goes off the rails? Microsoft wraps Scout in guardrails: its own governed identity, access scoped to the task at hand and stripped from the logs, human sign-off required for sensitive actions. These precautions say one thing: nobody is letting go of the reins just yet. The promised autonomy stays under watch, and the watcher is us.

The personal agent is no longer a parlor hypothesis. It comes in through the doors we leave open, the calendar, the inbox, the phone, and it promises a rare luxury: no longer having to think of everything. The contest playing out this summer is not over one more gadget, but over trust: how far to delegate the care of your own life to software that, to manage it well, must first know it whole. The answer will not be found on a spec sheet. It will be decided one consent at a time, in the quiet of our mornings.