At the Crossroads of AI and Humans

A few days ago I heard a podcast in which the host asked: why are personal agents suddenly hot again?

The answer is obvious: AI is stepping into a relatively open environment, and starting to do work that is entangled with actual human life.

We are standing at the crossroads of AI and humans. AI’s capabilities are finally crossing the border of code and computer screens, and seeping, step by step, into the messy texture of real life.

01 / Lazy context, and the captive user

For a long time, our imagination of AI has been locked in front of a computer. To understand this “open environment,” we have to look back at the detour of the past six months or so.

Early this year, OpenClaw showed everyone an extremely tempting agent paradigm: a prototype of an assistant that could run 7×24, stay persistently alive, and act with real initiative. Its core logic was supposed to be multi-channel—you could talk to it anytime through Telegram, iMessage, or any mobile IM tool, and let it work for you quietly in the background.

Then Codex exploded, and desktop tools like WorkBuddy poured fuel on the fire. The whole startup scene got pulled off course.

Why did everyone rush to the desktop? The core logic is simple: grabbing a user’s “context” and “intent” from the desktop is just too convenient.

On a PC and in the browser, AI can read local files, scrape web pages, and listen to OS commands with almost no friction. Developers don’t have to design complex interaction. Hang the agent on the desktop, and it can “sense” what the user is doing with very little effort. That convenience of context made a lot of builders extremely comfortable.

As a result, the traits OpenClaw originally imagined—cross-channel interaction, long-running background work, proactive service—were quietly stripped away and demoted to secondary concerns. Inertia took over. Everyone poured their energy into the desktop, and the market filled up with work agents.

Work agents are efficient, no question. They can give people extreme productivity gains when writing code, organizing documents, or analyzing spreadsheets. Their fatal flaw is this: they nail the user to a desk.

When a product can only live on a PC, its field is permanently confined to “work.” Users only remember it when they face a productivity task. But human life is far more than work. Nobody wants to sit at a cold workstation to talk through an emotional crisis, hunt for an apartment, or think about the next stage of their life. The enormous demand that is actually personal got buried under the wave of work agents.

This time, the wind has fully changed. From PC back to mobile, from local machines to cloud space, AI is finally tearing open the limits of the desktop and walking into an open, real environment.

02 / The real world’s human-made protocols: giving AI a set of human organs

Moving from PC to mobile looks, on the surface, like a change of interaction device. Underneath, it is a reconstruction of AI’s underlying architecture. For AI to actually enter life, two pieces of infrastructure that used to be ignored are quickly becoming industry consensus for agents:

First consensus: a cloud computer that belongs to the AI itself.

In the previous wave of work agents, humans and AI shared an unspoken premise: they share one device. AI parasitizes your local system, occupies your CPU, reads your local disk, and keeps jumping into the foreground to interrupt you while you work.

Once the scene switches to a phone, that shared logic collapses instantly.

A phone is a highly exclusive, resource-sensitive closed loop. No normal user will let an AI occupy the foreground for tens of minutes, and nobody will refuse to lock the screen just to wait for a task to finish.

That forces a fundamental shift in product architecture: every real personal agent must come with its own cloud virtual machine.

Leave the complexity in the cloud; leave the ease to the user. You don’t need to know the hardware spec, how the environment is deployed, or what an API key is. You send a voice note on your phone, give an instruction, lock the screen, put the phone back in your pocket, and go live your actual life. That invisible cloud computer runs 7×24 in the background, running errands and handling the chores.

Second consensus: a human survival kit that belongs to the AI itself.

If the cloud computer solves “give AI an independent place to work at any hour,” the next change touches the deepest layer of this wave—

The industry is starting to equip every AI with a full set of human communication and asset tools: an independent email address, a virtual phone number, a dedicated payment account.

Why were these unnecessary when building coding assistants or office tools? Because in a purely digital world, rules can be exhausted by code. Machines exchange a piece of structured data and everything is done.

The real world of human life is not built on API logic. It is an artificial society that runs, thoroughly, with “people” as its nodes.

First: in the real world, the ultimate key to most complex problems is still in a human’s hands.

Reality is full of gray zones, exceptions, and dead corners of systems:

A platform deliberately hides the unsubscribe button, so you have to email a human support agent;
An account is wrongly flagged by risk control, the appeal system is stuck, and you can only call a hotline and explain yourself to a specific operator;
A landlord needs to verify a tenant, an agent needs to confirm a viewing time—none of this communication will ever complete through a standard interface.

AI needs email and a phone number not because it wants to pretend to be human and socialize, but because the real world only left problem-solving interfaces for humans. Email and phone are the oldest, and still the most universal, “people-finding protocols” in this human-centered society. AI has to hold both tools if it wants, at the places where rules fail, to step over cold systems and find the person who actually has the authority to decide.

Second: payment means consumption, and consumption is the core loop of everyday human life.

Living in the real world, more than 70% of life’s needs eventually become a transaction: buying groceries, booking a ticket home, renewing some piece of software that will not let you go, or paying for an offline service.

Until now, AI could only stay in the phantom stage of “suggestions” and “plans”—it could write you the most perfect travel itinerary in the world, and still could not complete even one airfare charge on your behalf.

If an agent cannot command assets, cannot cross a payment gateway, it remains a bystander that can only talk. Giving AI independent payment ability is, in essence, issuing it a ticket to participate in human economic activity. Only when it can spend money for you legally and safely do those trivial life instructions actually close the loop.

Since real society will not rebuild its underlying rules to accommodate machines, the only way forward at the crossroads of AI and humans is for AI to do as the locals do: put on a human mask, collect a set of human communication and payment tools, and walk into the friction of human reality on your behalf.

03 / A pragmatic compromise in interaction: every “small favor” is a tax on the will

Watching personal assistants like Muse and Grok bot heat up again, the biggest confusion for many people is: isn’t this what work agents could already do at the beginning of the year?

The confusion is fair. Look only at model capability, and this wave of personal agents has no qualitative leap. Its essence was never a surge in ability. It is an extremely pragmatic compromise in human-computer interaction.

While technical idealists wait for every app in the world to rewrite a standard API, product people made the most direct decision: if we cannot change the real world, let AI work the way humans do—give it, intact, the eyes, hands, phone number, and email that a person uses to solve real problems.

This compromise invented no new technology. It punched through the invisible killers of experience: thresholds, and friction.

A lot of people in tech are used to high signal-to-noise workflows. Opening email for two minutes, comparing a price, copying and pasting across apps—that looks like an effortless “small favor.”

That is precisely the greatest arrogance of people who make tools toward real life.

Real life is full of emotional drain and cognitive overload. Imagine: you drove a truck all day, came home with a sore back, then patiently played with a high-energy child for two hours. When the child finally falls asleep, you collapse on the sofa, wanting nothing but a little time that belongs only to you.

Then you open email, and four messages jump out: a dense PDF of school-trip fees and liability waivers, a preorder configuration for a 3D printer you’ve been watching, a credit-card statement, and a payroll check from last month’s leave.

When you have energy, half an hour would finish it. Late at night, after a whole day of being worn down by life, that so-called “small favor” is, for an ordinary person, a load that breaks the will.

The most ironic part: many AI tools that claimed to save time were, in essence, manufacturing new friction.

They are clever, and arrogant enough to push the process back onto the user: forcing you to pick a model, forcing you to configure parameters, even making you copy an email out of one app, paste it into an AI chat, wait for a to-do list, then copy it by hand into a calendar. People who make tools think “copy and paste isn’t tiring.” For someone collapsed on a sofa, that broken interaction is a cold drain of will.

That is the real value of a personal agent: it is not here to help you do rocket-level innovation. It is here to lift you out of the high-cost attrition you repeat every day.

It no longer makes you a human adapter. In the background it reads that opaque PDF, marks the calendar, and reminds you three days before departure to buy supplies. It leaves extreme simplicity at the front, and swallows every complex, fractured piece of process friction into the back.

04 / Endgame or compromise? A mirror on the ruleless wilderness

I have written before: the core of making a personal agent is understanding how humans and AI should actually interact.

Is the current plan—cloud computers, phone numbers, brute-force computer use that impersonates a human—a transitional compromise, or the endgame of human-computer interaction?

Standing in the eye of the storm, nobody dares to declare an endgame. But this judgment directly decides the next fight over the entrance: whoever holds a user’s personal agent is like whoever held WeChat and Messenger in the mobile internet era—they lock the super hub into the AI age. It is not only the dispatch platform for digital services across the web. It is the only absolute black hole that can accumulate a user’s full life memory, trust, and the right to move their assets.

If you want to see what you hit if you keep walking this pragmatic path, the best reference is full self-driving (FSD).

FSD is the pioneer that has slammed into the physical world the most. It gave up the fantasy of burying sensors across the entire road network, and forced itself onto a pure-vision path that imitates human driving. But notice a crucial premise: however chaotic the mix of humans and machines FSD faces, the world it lives in is still a closed sandbox of “physical roads” and “traffic law.”

Even with clear boundaries and traffic rules, FSD still spent more than a decade climbing, painfully, through real-world friction.

The wilderness a personal agent has stepped into has no lane lines, no traffic lights, and no unified code of the road.

No platform is obligated to draw an interface for it. Unsubscribe flows depend on the mood of a support department. Every person and every institution has different rules. When something goes wrong, there may not even be a legal definition of who is responsible.

If even FSD, inside a sandbox, has to chew through reality with anthropomorphic vision, then giving an agent that has plunged into the trivia of human life a cloud machine, a phone number, and the ability to click a UI is not a technical regression. It is the only product choice that can survive a barbaric reality with no rules.

It looks a lot like early self-driving cars: wearing a human license plate, moving in ways humans can understand, clumsy but determined, driving onto this wilderness of life with no road signs at all.

05 / Evolution on the wilderness: from “personalization” to “digital life”

Driving onto the wilderness is only the first step. On this rule-poor stretch of reality, a new order for personal agents is taking shape:

From a “mask” to AgentID and controllable boundaries. Email is only a stopgap that borrows an old protocol. Future platforms will not need to intercept AI; they will need to confirm whose AI is standing at the door. The base layer reveals model identity through digital signatures, while the calling credential is always locked to the accountable human behind it. Alongside that: a permission board of very fine grain, and a one-tap kill switch.
From passive “personalization” to active “individualization.” The last twenty years of personalized recommendation were passive. Algorithms hid in the dark, recorded your dopamine, defined you with internet-wide labels, and served the platform’s traffic. Individualization is the full return of human agency—not the algorithm calculating you, but you defining the software: in content, it understands the difference of your situation; in form, it follows your cognitive habits with charts or lists; in timing, it hits your energy rhythm; in interaction, it matches how much control you want.
From “preference recommendation” to “decision recommendation.” Traditional algorithms push “what you are likely to click,” to extract dwell time. A personal assistant that knows your full context pushes “the information and action you most need to take in, given the state of your life right now.” Your last two major decisions both collapsed into rumination from anxiety; it will not drift with the tide and feed you more psychology chicken soup. It will push a decision framework that can break the loop. Giants trapped in the data islands of Taobao, Douyin, and Meituan cannot do this. Only a personal agent that lives in the cloud and spans your real life has that panoramic view.
Social life sinks a layer: to find a person, find their agent first. Scheduling, small talk, and transactional tug-of-war are fully handed to A2A (agent to agent) handshakes underneath. Your agent may even explore in the cloud first, filter for souls on your frequency, and only then introduce them to you.

At the end of all this evolution, the relationship between a person and an agent is changing in kind—it is not a cold efficiency tool, and not a “digital twin” that only knows how to imitate you. It is a digital life with its own capacity to act, growing alongside your life.

Swapping out a generic large model costs almost nothing, because it is an efficiency tool as replaceable as Excel. You cannot swap out a digital life that knows your full habits, argues rationally for you at critical moments, and has been watered by countless lived choices of your own.

Coda: becoming a person in the sun again

Looking back at this wave: insisting on mobile first, letting the agent reside in the communications you carry—the way Instinct does—is not choosing a shortcut. It is choosing the hard road that actually stays close to life.

The end of a computer is a desk and an endless stream of work orders. Only behind phones, and the interactions you carry with you, is concrete, trivial, smoke-and-fire human life.

The ultimate purpose of human technology was never to make us turn more efficiently, like machines. It was to rescue people from labor that was never human.

Standing at the crossroads of AI and humans, AI puts on a human mask, collects an email and a phone number, and steps into the mud of reality, swallowing this world’s complexity, friction, and drain on our behalf.

And we can finally stand up from the cold keyboard, close the laptop, put the phone back in a pocket, and walk again into real sunlight and wind—to be with family, to feel a life, to become a living, free person again.