AI can generate a hundred different lives, yet it cannot grant you one extra fallback option

5 days, 15 million life plans.

This June, Qianwen launched its college entrance exam application Agent. Input scores, rankings and a few preferences, and a fully actionable plan ready for direct submission is generated. Within 5 days of launch, it produced 15 million application reports. Tencent, Baidu and ByteDance are also developing similar products, most of which are free of charge.

An examinee can explore dozens of possible futures in half an hour: staying in their hometown province to study medicine, moving to another city to pursue computer science, choosing a university with guaranteed postgraduate recommendation opportunities, or giving up a bit of major ranking for a beloved city. Each option comes with supporting data, clear reasoning, and a seemingly complete blueprint for the future.

When it comes to actually submitting the application, things suddenly shrink to tiny, specific details. The father asks about the city, the mother worries about employment prospects, and the child stares at a university they have never visited, repeatedly asking the same question: What if I finish one year and realize I don't like it?

AI can regenerate a new plan in a single second, but the whole family must stop debating before the deadline arrives.

15 million reports did not create a single extra university admission spot. 15 million new paths appeared on the screen, but not a single additional seat materialized in reality.

Free Agents have not eliminated paid consulting services. One-on-one services at an agency in Chengdu still sell for up to 12,980 yuan, with all slots fully booked; a local career planner admits that AI has taken over the work of researching information and comparing data, so parents spend less time in face-to-face discussions, yet end up paying no less than before.

Answers are all free now, so why are parents still willing to spend over ten thousand yuan?

Perhaps what they are buying is no longer more answers, but someone finally telling the whole family: That's it, stop looking.

In the past, technology solved the problem of having too few answers; now, it has created the opposite problem: an overabundance of possibilities.

AI has brought the "what if" into the present

In 1986, psychologists Markus and Nurius proposed the concept of "possible selves": people do not only live in the "current self", but are constantly driven by "who I might become", "who I hope to become" and "who I fear to become".

In the past, forming a "possible self" was costly. To imagine switching to another industry, you first had to know people working in that field; to determine if studying abroad was feasible, you needed to research universities, calculate expenses, and consult seniors. Most un-traveled paths were nothing more than vague thoughts. Because they were unclear, people could easily let them go.

Generative AI has changed this. Submit your resume, savings, desired city and family situation, and it will immediately generate a complete career-switch roadmap; after a few follow-up inquiries, it will supplement the plan with course schedules, portfolio guidelines, target companies and a 90-day progress tracker. "Others might be able to do this" has now become "here is exactly how you can do it".

AI does not just generate answers — it creates a highly detailed alternative life that bears your name.

Late at night, people do not usually ask AI "what exactly should I do", but rather "if I start now, is it still too late?" AI almost always offers a path that makes you believe it is not too late. It does not judge you for being too old, does not remind you that your last plan only lasted two weeks, and does not advise you to give up even after you have failed three times. This ability to never let people lose hope is both gentle and dangerous.

This is not entirely an illusion. In 2024, a writing experiment published in *Science Advances* found that people assisted by AI produced work with significantly higher scores, and the improvement was most notable for those who previously had lower creativity levels; at the same time, however, these works grew increasingly similar to one another. Individual ideas became better, but overall diversity decreased. This experiment cannot directly explain life choices, but the trend is consistent: AI makes everyone feel they have more paths available, but when they look up, everyone is holding similar roadmaps — switching to become an AI product manager, building personal media brands, taking overseas orders, all crowding into the same set of positions and the same market. The roadmap is customized for each person, but the destination is growing more and more crowded.

I call this phenomenon "possibility inflation": the number of clearly describable life paths is growing far faster than people's actual ability to live them out.

Some might argue that too many choices cause distress, a point already made in popular psychology bestsellers 20 years ago. The difference lies in magnitude. Past options were vague "lives elsewhere" on a shelf; now, they are complete scripts tailored to your name, your savings, and a 90-day schedule. Old-fashioned choice overload made people dizzy from too many options, while new possibility inflation turns every untraveled path into a precise, outstanding debt.

In the past, the scarce resource was "are there other paths available". Now, the scarce resource is deciding which paths to close off among all the seemingly viable options.

Some advice is not useless — you just cannot afford it

The most common misunderstanding AI creates is disguising "describable" as "achievable".

A college application report can clearly list safe, reach and backup options, but it cannot create extra admission spots; a job-hunting plan can polish your resume, but cannot conjure up a new position; a startup business plan can complete market analysis and financial models, but cannot automatically bring in your first paying customer. This is not because AI is not powerful enough — information and opportunities are inherently two different things.

Last September, *The Atlantic* reported on Harris, a graduate from the University of California, Davis: he used ChatGPT to refine his application materials, submitted roughly 200 job applications, and received no offers — many companies did not even send rejection letters. On the other side, enterprises are also using AI to write job descriptions, screen candidates and arrange interviews. The cost for job seekers to submit applications has dropped, and the cost for employers to reject candidates has also dropped, but the number of positions does not increase just because there are more resumes. When both sides use AI, a more complete resume quickly stops being an advantage and becomes a basic admission ticket, pushing competition further back: real project experience, internal referrals, on-site interview performance, and how long you can hold out.

Using financial terminology, AI is distributing large quantities of "paper options": it tells you there is a path ahead that you are qualified to take. But to convert the option into something real, you have to pay the strike price: time, savings, health, family support, and the ability to start over after failure. AI can generate a free 6-month career-switch plan, but it will not receive a landlord's rent notice at the end of the month, nor will it have to explain to parents why you are starting over as an intern after three years of working.

Therefore, what AI popularizes first is not opportunities, but the visibility of opportunities: it lets more people see the door, but does not prepare a ticket for anyone. Seeing the door certainly has value, but for people who cannot afford the ticket price, a clearer path only brings a more precise sense of powerlessness.

There are a hundred paths on the screen, but you only have one pair of feet in reality

The fundamental difference between machines and humans is not who calculates faster — it is that AI can run through a hundred drafts of your life, but you only get to submit one final version.

It can open a hundred conversations simultaneously, simulate a hundred different careers, and reset everything at any time. It has no age, no rent to pay, and will not truly lose the years it could have spent in Chengdu just because it "chose" Beijing. Humans only have one body, a limited amount of time, and one personal balance sheet. Real choices always mean gaining one thing while giving up several others. The true cost of a path is often not written on the path itself — it is written on all the paths you can no longer take.

AI has made those abandoned paths more concrete than ever before. When work is frustrating, it can immediately draft a business plan for opening a café; if your personal media account fails to gain traction after three months, it can generate a complete cross-border e-commerce roadmap. Every plan sounds reasonable, and every one feels more complete than your current stumbling life — because none of them have paid the cost of failure yet.

When generating a new path becomes cheaper than fixing an old one, plans shift from being action tools to consumer products. This is why we love asking AI to rewrite plans: not because the previous version was truly unworkable, but because the feeling of starting over is always more comfortable than admitting you could not stick to the original plan.

This has spawned a new form of procrastination that looks extremely productive: discussing positioning with the model, comparing ten different plans, generating three versions of schedules, and redesigning your entire life from scratch. Documents grow more and more complete, but the actual time you invest in any single path becomes increasingly fragmented. More subtly, the very act of generating a plan creates the comforting illusion that "I have already started". The moment you close the chat window, you might even sigh in relief, as if a more disciplined, braver version of you has already moved forward a few steps. But when you wake up the next day, your life is still exactly where it was.

AI excels at opening new branches, but human growth requires closing them off. Mastering a craft, running a business, building a professional reputation — all rely on long-term repetition, demanding that you give up other temptations before seeing results. Commitment, at its core, is actively limiting your future: it is not that you cannot see other paths, but that you see them and still decide not to take them. This sounds unwise, even a little foolish. Smart people can always spot new opportunities, but the real challenge is to see them and still go back to the unfinished task you were working on yesterday, even with no immediate feedback.

This is exactly why that sentence "That's it, stop looking" is so valuable. What parents buy is not that the consultant knows more universities than AI, but that someone is willing to end the search for the whole family, compressing all the information into a decision that will not be second-guessed.

The cheaper answers become, the more valuable services that help people close off possibilities will be.

Do not expect generative systems to voluntarily advise you to stop. Commercial products naturally want you to use them more frequently and for longer; generative systems are designed to answer "are there other possibilities", but rarely tell you after the tenth plan: enough, the remaining questions can only be answered by living your life. When opening new branches aligns better with product logic than closing them, possibility inflation is not just a psychological phenomenon — it has commercial forces fueling it.

The same plan is for trial and error for some, and unaffordable for others

To be fair to AI, not all the possibilities it distributes exist only on paper.

A 2025 study published in the *Quarterly Journal of Economics* tracked 5,172 customer service representatives: after adopting generative AI, their per capita productivity increased by roughly 15%, and the employees who benefited the most were exactly those with the least experience. These tasks share common traits: clear objectives, feedback arriving within minutes, and the ability to correct mistakes on the next call. In these well-defined, fast-feedback scenarios, AI delivers tangible leveling effects — but it levels productivity, not the subsequent income and social status.

But education, career, marriage, relocation, and entrepreneurship are not this type of task. Their feedback arrives years later, their outcomes are mutually exclusive, and failure cannot be undone with a single click. Choosing a customer service script and choosing a city can both be done with AI's help, but the cost of making a wrong decision is not on the same scale.

This is why "AI helps ordinary people" and "AI does not necessarily make life opportunities more equal" can both be true. Economist Amartya Sen distinguished between resources and substantive freedom: not everyone can convert the same resource into the same quality of life. The same AI account, in the hands of someone with savings, free time, and family support, becomes a set of low-cost testable projects; in the hands of someone who must make loan repayments on time next month, it becomes a set of unexercisable suggestions.

Therefore, the truly critical personal asset in the AI era is not just cognition, but "trial-and-error capital": how much time you can invest to test suggestions, how many failures you can withstand, whether you can keep investing even when there is no short-term return, and if you have a fallback position after losing. Some people can test a new suggestion every week: build a simple webpage, spend a few hundred yuan on advertising, meet three industry peers for conversations, and switch to something else if it does not work — failure only eliminates one option. Others finish work at 10 PM, need to take care of their children on weekends, and do not have enough savings to take a month off work. They are not incapable of understanding the roadmap AI provided — that roadmap requires them to already live a completely different life. The first group treats a hundred possibilities as experiments, while the second group receives a hundred perfectly correct plans, none of which they can afford to take on.

This is what "fallback position" in the title means. A fallback position is not passive security — it is the infrastructure for action: because you can afford to fail, you dare to start. AI can push the cognitive threshold to the floor, but it cannot create this safety buffer.

Agents click buttons for you, but you still have to live the rest of your life

Moving forward, AI will naturally continue to advance toward the action layer. College application Agents can already directly generate a fully submittable application form, job-hunting Agents submit resumes in batches, and startup Agents build websites, send emails, and analyze data. The distance between an intention and actual operation will keep shrinking.

But the rule separating drafts from final versions has not changed. An Agent can submit 200 resumes on your behalf, but cannot work full three years at the new job for you; it can generate an application form, but cannot complete your university degree for you; it can contact a hundred clients, but cannot fulfill the commitments to ten of them. It clicks many buttons for you, but the rest of your life is still yours to live.

Following this logic, what the next generation of AI products should truly compete on may not be who can generate the 101st plan, but who dares to ask you when you say you want to quit your job for the tenth time: Did you run that small experiment from the last plan?

A good AI should not only be responsible for opening new paths — it should also help you wrap things up: explicitly eliminate options that require risks you cannot afford or resources you do not have; replace the grand script of "quitting your job to switch careers" with a two-week experiment you can run without leaving your current role; remember why you started and what you have already invested, and remind you to finish the task at hand when you are tempted to start all over again. The better AI gets at opening branches, the more we need systems that help us close them. This also redefines the value of consultants, coaches, and organizations: now that knowledge is no longer scarce, what they sell is no longer "I know the answer" — it is commitment, coordination, and shared responsibility.

AI makes all the unlived lives come back to you

Every technology rewrites how people interpret regret.

In eras of information scarcity, not traveling far or switching careers could be attributed to "I did not know this path existed". Unlived lives never clearly appeared, so they never came back to haunt you every day.

AI has taken that ambiguity away. It will tell you exactly what to learn starting today, what positions you can apply for in six months, and what you can achieve by devoting two hours a day for a year. That unchosen path is no longer a vague "maybe" — it is a complete project plan with specific dates, budgets, and step-by-step instructions.

When society mistakes visibility for opportunity, the blame for failure will increasingly fall on the individual. The roadmap is given to you, courses are free, tools are operated for you — if you do not reach the destination, there seems to be only one explanation: you did not try hard enough. But a roadmap is not a real road, and knowing how to exercise an option does not mean you can afford the strike price. When a person fails to live up to a certain possibility, it is usually not because they could not understand the plan — it is because their time, assets, health, and responsibilities only allow them to live one single life.

What AI truly brings is not just the democratization of knowledge, but the industrialization of "possible selves". For the first time, we can preview all these unlived lives at such a low cost, yet we still only get one lifetime to live out one of them.

One late night at the end of June, an examinee who has explored a hundred different futures finally presses the submit button. At that exact moment, the other 99 possible lives are not proven wrong — they simply no longer belong to them