The Helpful Agent Wants Your Inbox, Your Calendar and Your Bank Account
Somewhere in the last year, AI agents got a promotion. They no longer wait for you to ask them something. They brief you in the morning, watch your inbox, book your flights, and — if you let them — do your shopping. Meta, OpenAI, Anthropic and xAI are now going head-to-head selling you this creature, the consumer AI agent, and the pitch is seductive: a colleague who never sleeps, or a personal assistant who never forgets.
There is only one problem, and it is the whole story. The agent gets better the more it knows about you. And for the tasks that matter — sorting your email, booking your flights, doing your shopping — the access it treats as par for the course runs deep: inbox, calendar, identification, bank account.
There is no holy grail. There is a stack.
His answer — and that of many others who use AI agents regularly, according to the AP’s reporting — is to keep a stack: a combination of agents from different labs, each doing what it is good at, and — in a twist that would make any manager smile — checking each other’s work.
“What I like to do is to have this system of checks and balances in place where one agent does the work and then I let the other one critique and criticize it, and that way, they find mistakes, correct themselves and become better,” said Clara Gigling, founder of Oudae Labs.
So the future of personal computing, at least this year, is not one brilliant servant. It is a committee. And committees, as any office worker knows, need managing.
Meet the committee of AI agents
Each of the big labs markets its agent a little differently, and the job titles they have chosen are revealing. Anthropic said it built Claude Duet for “professionals and small business owners with more on their plates than they can track.” Meta said its agent, Muse, can help with tasks from “saving money and improving your health to smart shopping and helping you strengthen your relationships.”
Claude Duet, OpenAI’s Dots and xAI’s Grok Bot have each been described as aiming to be like a colleague, or even a chief of staff. Muse and Instinct, from the startup Spear Street Technology, have been described as more like personal assistants. The distinction is real, but the AP’s reporting notes that none of them is incapable of helping outside its marketed realm — and the descriptions themselves show how each company positions its agent.
Artigas has his own seating plan. Claude Duet, he said, is the one for a side hustle or launching a new business. Dots is the debugger and problem solver for specialised technical work. Instinct handles the tedious stuff: emails, reservations, travel plans. And for a complete novice, he recommends Muse, because it is “very user-friendly.”
Some of these agents are free. Some require a monthly subscription. Others are still in testing or not available to everyone — Claude Duet, for instance, opened as a research preview in October, on a waitlist for US Pro and Max subscribers — and availability is restricted in countries outside the US. Before you hire the committee, check whether it is even hiring where you live.
The price of admission
Here is the part worth slowing down for. For an agent to sort through your deluge of emails, book your flight or do some holiday shopping, access to your inbox, your calendar, your personal identification information, your credit card number and your bank account is par for the course. The agents perform better the more they know about a person. That is not a bug in the design. It is the design.
Gigling’s advice is to be curious, but careful. “Curiosity is always a great starting point, just trying and playing around with it,” she said — “but it’s also important to be mindful when sharing private information.” The companies, for their part, have worked to address those concerns: privacy policies that describe agents running on their own dedicated cloud computers, and controls over what the agent remembers. It is worth saying plainly that this bargain is not new: email and search already hold much of this same data, and it is possible that an agent on a dedicated cloud computer, with approval gates, is safer than the services we already trust with our inboxes — the evidence does not settle that comparison either way.
Some of those safeguards are genuinely thoughtful. Anthropic says Duet runs on its own cloud computer with its own browser, carries over the memory, skills and connected apps from your Claude account, and must show you its intended step and get your approval before it sends a message, makes a payment, signs a document or deletes anything. Trading securities and moving money between accounts are off-limits even with approval. Duet never directly sees passwords or sign-in tokens, and for purchases it uses a one-time card, so the real card number is not shared. A classifier is designed to block it from sharing credit card numbers or government IDs at all.
Meta’s Muse runs in a dedicated cloud computer of its own, and Meta says it does not share your conversations or the data in your virtual machine with its ad systems. But there are catches the fine print does not hide, exactly. Meta’s own privacy policy says the setting that lets your interactions train its AI models is on when you first use Muse — you have to switch it off. And Meta warns that after you delete something, Muse may still “remember” information it learned from it; only Reset Muse wipes everything permanently. The stronger fix Meta announced — a Confidential VM, encrypted with a key only you hold, “so not even Meta can access it” — was still described as coming in the future as of early November.
Start with one annoying thing
So how do you actually begin? The experts’ advice is refreshingly unglamorous. “Start with one annoying, repeated thing that you keep putting off,” Artigas said — with the caveat that it has to be an online task. An agent cannot fold the pile of laundry or organise the junk drawer. A long email you do not want to write, or a flight you have been slow to book, are good jumping-off points.
Be specific, Gigling adds: give the agent a detailed goal, the required context and the tools to achieve it. Then, as Artigas puts it, “You just have to experience the different agents and see what works best” for you.
That is honest advice, and it is also the quiet confession at the heart of this whole market. Nobody can tell you which agent fits, because the fit depends on the work — and on the bargain. The useful agents are the ones that ask for the keys: your inbox, your calendar, your identification, your bank account. The companies have built approval gates, one-time cards and memory controls to make that bargain safer. But the bargain is still yours to make, one agent at a time, and the real question was never which one is smartest. It is how much of yourself you are willing to hand over — and to how many of them.
