After AI makes answers cheap, what is really scarce is a sense of direction: reading Superagency

When I talk to friends about AI recently, the conversation often quickly falls into two paths. One kind of person is excited to show off new tools (opens in a new tab), telling you that one afternoon can complete the work of the past week; the other kind of person is worried about work being replaced, false information getting out of control, and eventually even human judgment being outsourced.

Superagency (opens in a new tab), co-authored by Reid Hoffman and Greg Beato, does not stop at these two familiar positions. It raises a more worthy question: If AI gives everyone access to the wisdom that only a few experts once had, what will we do with this ability?

After reading this, my biggest gain is this sentence: What is truly scarce in the AI ​​era may no longer be answers, but the ability to act autonomously. Tools can provide suggestions, organize information, and even perform some tasks for us, but a person’s ability to set directions, judge results, and bear choices determines where these abilities will ultimately lead.

The changes brought about by AI cannot be measured by efficiency alone

The book compares the “https://zh.wikipedia.org/zh-tw/%E5%90%88%E6%88%90%E6%99%BA%E8%83%BD (opens in a new tab))” (synthetic intelligence) provided by AI to the “synthetic energy” of the industrial revolution. Steam engines allow humans to expand production scale without relying solely on human and animal power; AI makes wisdom a resource that can be called upon, expanded, and continuously strengthened.

This analogy made me rethink the way I use AI. Many people come into contact with AI for the first time, and the most common question they ask is: “How much time can it save me?” So we use it to write letters, organize meeting minutes, and generate posts. There is nothing wrong with these uses, but if we think about it calmly, we still think of AI as a faster machine.

If wisdom has become a resource that can be used flexibly, a better question might be: “Is there an opportunity now to start something that I was afraid to do in the past because of lack of knowledge, time, or expertise?”

A small entrepreneur can understand a contract faster, a teacher can design teaching materials according to the differences of students, and a patient can sort out symptoms and questions before seeking medical treatment. What these things have in common is that what originally stuck them was not the execution, but the threshold. What AI really amplifies is the possibility of people being willing to take the next step.

This is why I have repeatedly emphasized that questioning power is more valuable than prompt words. When execution becomes cheaper, the decisive factor moves forward to “what problem are you trying to solve?”

Personal becoming stronger is just the starting point of the super agent effect

What’s most interesting about Superagency is that it doesn’t focus on a handful of super users. When more and more people use AI to improve their judgment and action, these abilities will multiply with each other through the connections between people. In the end, even people who do not directly use AI may also benefit.

The book cites an easy-to-understand scenario: Even if the patient does not use AI, as long as doctors can identify vague and seemingly unrelated symptoms through AI, the patient will still receive better care. If car technicians can find out the cause of engine noise more quickly, car owners will also benefit.

This also reminds me that when measuring the value of AI, you can’t just count the hours you save. What’s more worth observing is: does this freed-up time improve team collaboration, customer service, or public decision-making?

If everyone just produces more mediocre content faster, efficiency will appear to be improved at first glance, but it will not help improve the overall environment. Individual empowerment must be connected and shared before it has the chance to become collective progress.

You don’t have to wait for perfection, but you can’t stop verifying

The book advocates “iterative deployment”: allowing AI that is not yet perfect to enter the real world, and gradually become safe and reliable through use, feedback, and corrections.

This view seems to me to be pragmatic. If technology remains in the laboratory, people will have no way of knowing what mistakes it will make in real situations, and it will be difficult to establish specific regulations. I have seen the same kind of failure too many times at Enterprises introducing AI: they waited until the solution was perfect before daring to launch it. As a result, when it was launched, no one remembered what problem it was trying to solve.

But this optimism must be paired with responsibility. Making corrections as we move forward does not mean we have to wait until something goes wrong. In high-risk areas such as medical care, finance, education and public governance, it is necessary to clearly mark the degree of AI participation, retain human review, and allow affected people to have the right to appeal and refuse. This is exactly the same position I took when I talked about disclosure of AI use in paper submissions: Disclosure is not about exposing one’s own shortcomings, but about laying the foundation of trust on the table.

After all, trust doesn’t happen automatically just because AI gets smarter. Trust comes from traceable processes, open limits, and someone willing to take responsibility for the results. If these conditions are not met, “super agent” may become a beautiful term for a few platforms to expand their power.

In the AI era, practice becoming a person with direction first

What “Super Agent Effect” brought to me was not a slogan to “embrace AI”, but a more serious reminder: when the cost of acquiring wisdom drops rapidly, people’s sense of direction will become more important.

This can be practiced starting with a real task. Don’t just ask the AI ​​to give you an answer. Instead, state clearly what you want to improve, then ask it to provide options and point out risks, and then make the final judgment yourself. After completion, record which links are worth handing over to AI and which links must retain human experience and responsibility.

The reason why this exercise is important is that it cultivates two things at the same time: first, the ability to explain the task clearly, which is the premise of Leave the job to the AI ​​Legion; second, prevent the thinking muscle from degenerating. I mentioned when talking about Brain-protecting AI learning method that AI will not make you stupid, but it will not automatically make you stronger. The difference lies in whether you leave your own judgment.

Incidentally, Hoffman himself is the co-founder of LinkedIn. If you want to look one level down, the [5C Framework](https://www.vista.tw/blog/5c-framework-ai-era proposed by LinkedIn CEO talks about the other half of the same thing: the five things that AI cannot do are exactly where people should put their efforts.

Whether AI will make society a better place has yet to be determined. The pragmatic optimism proposed by Hoffman is not the belief that technology will naturally bring about progress, but the belief that people can shape technology through participation, verification and modification.

When wisdom is as ubiquitous as energy (opens in a new tab), the future gap will not only depend on who has a stronger language model, but also on who knows better where they want to go and is willing to let the amplified ability benefit more people.

I think this may be the most valuable lesson for us to think about in Superagency.