What competitive advantages are still defensible in the AI era?

Original Title: How To Reason About A Messy Future
Original Author: Systematic Long Short
Translation: Peggy, BlockBeats

Editor's Note: As AI begins to write code, optimize code, and even gradually take over the software production process, a deeper structural change is looming: division of labor, corporate organization, and even knowledge barriers may be redefined.

The author of this article was once part of a nearly 20-person team at a hedge fund but chose to leave for entrepreneurship during his career advancement. In his view, the real signal is not market sentiment, but the leap in technical ability. When models can consistently generate usable code and possess recursive improvement capability, the logic of software development and knowledge production has already begun to change.

The article, from the perspective of quantitative finance, analyzes several types of short-term "moats" that may still exist in the AI era, including proprietary data, regulatory friction, authority endorsement, and lag in the physical world. It also puts forward a core judgment: in a highly uncertain era, more important than accurately predicting the future is identifying directions and taking action before the window closes.

Below is the original article:

When Models Start Writing Code, Change Becomes Irreversible

The first time I realized the industry was approaching a turning point was in my previous job. It felt like hearing the background music slow down while everyone around me was still pretending that nothing would change.

At that time, I was managing a team of nearly 20 people at a hedge fund, doing what I had been doing for many years. From the outside, this was almost a steadily rising career path. If I had stayed there, I would probably have achieved even greater success. However, in the end, I chose to leave that position that many people dreamt of and start a new venture from scratch with a team of only a few people. This decision was almost incomprehensible at the time and was even seen as a form of "career suicide."

But in recent months, massive layoffs, voluntary entrepreneurship after leaving jobs, and more and more people working during the day while quietly coding and working on projects at night. All of this has made that decision that seemed "crazy" at the time seem less far-fetched.

During this time, many people have asked me: where will all this ultimately lead? This article is the answer I can currently provide.

Frankly, I am not sure how significant the change will ultimately be. But one thing quantitative finance has taught me is: being on the right path is often sufficient.

What truly made me realize that the change was irreversible was the ChatGPT o1 model.

Before that, I always referred to these systems as "LLMs," not "AIs." I didn't think they really had any sort of intelligence-like capability. But when o1 came around, something changed: these models could, for the first time, stably generate code through structured prompts.

The code was still imperfect and could suffer from hallucinations or misunderstandings. But the key was this: it could now write useful code.

My judgment was simple. Once AI could generate usable code, it would recursively start improving its logic and driving software development at a speed we can hardly imagine.

Every time I make this point, someone always argues, "This code still has bugs and is far from meeting production standards." But this overlooks a fact: human-written code also has bugs. We don't need AI to write perfect code in order to stop writing code ourselves.

The real turning point is when the error rate of AI-written code is lower than that of humans, while being much faster. At that moment, the act of writing code will be completely outsourced to machines.

After witnessing the capabilities of o1 firsthand, I can almost be certain: there will be very drastic changes in the future.

The Moat That Still Exists in the AI Era

Initially, I thought AI would gradually erode the quantitative finance industry, but this process would be relatively slow. The reason is simple: institutional-level code has almost no publicly available data for training.

At the time, I imagined software engineering as a pyramid: at the bottom was basic coding work; moving up was senior engineers with architecture capabilities; higher up were professional developers, such as data scientists, quantitative developers, and various industry experts. Theoretically, the deeper the expertise, the more secure the profession.

My initial assessment was that within two years, basic programmers would be the first to be eliminated; followed by senior engineers; further up, as the models gradually absorbed specialized knowledge, higher-level positions would also be impacted.

But soon I realized another thing: cutting-edge model companies would eventually directly hire industry experts to input specialized knowledge into the models. In other words, specialized knowledge would indeed be a short-term moat, but in the long term, it would also be gradually absorbed by the models.

In my assessment at the time, there were several types of businesses that were unlikely to be easily disrupted in the next five years.

Category One: Proprietary Data

Companies with a large amount of proprietary data are harder to replace.

For example, large multi-strategy hedge funds (pod shop), such as institutions like Millennium, generate massive amounts of data every day: analyst research, investment recommendations, market insights, actual trade results.

This data can be used to continuously refine models, creating a competitive advantage that is hard to replicate externally. As long as a company's data sources are not easily available to the model, it still maintains a certain time-based moat.

Category Two: Regulatory Friction

Any industry requiring significant human approval is not easily disrupted. For example, traditional financial markets.

To enter these markets, you need to: open a brokerage account, obtain licenses, sign cross-border legal documents. Trading crypto assets is easy, but a foreign company wanting to trade iron ore in China is far from simple.

As long as an industry still requires human signatures for approval, its pace of development will be constrained by approval processes.

Category Three: Authority as a Service

Now, having AI write a legal opinion is no longer a challenge. But the reality is that people are still willing to pay tens of thousands of dollars for a lawyer to provide legal advice. The reason is simple: AI's opinions currently lack authority.

The same logic applies to smart contract audits. Technically, AI may already match or even exceed top auditors' level. But the market still prefers to purchase the "stamp" of a well-known audit firm.

Because what customers are truly buying is not the opinion itself, but the authority behind the opinion.

Category Four: Physical World

The progress of hardware is much slower than software, and hardware issues are also more difficult to fix.

Therefore, industries that directly interact with the physical world are unlikely to be rapidly disrupted by AI in the short term. However, once hardware capabilities catch up, the same logic will still apply: lower-level positions will disappear first, followed by higher-level positions.

These moats do exist. But it must be acknowledged that they only delay change rather than stop it.

Act Based on Signals, Not Waiting for Certainty

When the future is highly uncertain and the pace of change is rapid, people often make two mistakes.

The first is waiting for certainty before acting. The second is simply applying historical analogies, such as: "This is like the dot-com bubble."

Both approaches can lead to judgment errors.

In situations of incomplete information, a more reasonable approach is to reason from first principles.

You don't need to know every detail of the future. You only need to roughly assess the direction, design asymmetric bets, meaning that if you judge wrongly, the loss is manageable; if you judge correctly, the gain is enormous.

In an uncertain future, asymmetry is everything.

A practical thinking method is to first ask yourself "What are the prerequisite conditions for a certain outcome to occur?" and then ask if these prerequisite conditions have already emerged?

In retrospect, this AI turning point was not difficult to foresee. Because the key inputs already existed: code that can write itself, models that can recursively improve, institutional knowledge that can be bought instead of nurtured.

As long as you carefully observe these signals, you can roughly assess the future direction.

You can even continue to extrapolate.

We may not have truly seen the following scenarios yet: AI that can train itself, AI that can replicate itself, AI that operates entirely autonomously.

If an AI can enhance its own capabilities by 0.1% through a series of actions, it may sound insignificant. But as long as this number is not 0, it will continue to amplify. This is a typical power law effect.

In the financial markets, once a signal becomes obvious, the trading is often already crowded.

In investment, you exchange uncertainty for early-stage belief. In both career and entrepreneurship, it is fundamentally the same.

So the real question is not, what will happen in the future? but rather, what do I already know? What direction do these pieces of information point to? What is the cost difference between acting now and waiting?

There is also a frequently overlooked fact that action itself creates information.

Action does not happen in a vacuum. When you take action in the world, the world provides feedback. This feedback brings new information. Information drives iteration. Iteration leads to better actions. This is the basic mechanism of progress.

Remaining still in uncertainty is a form of slow decay. Action, on the other hand, signifies exploration.

If I only want to continue to enjoy the dividends of the existing system, I may be able to maintain for a few more years. But I have always wanted to do something truly my own, and I feel like this window is closing rapidly.

Of course, the world's largest hedge funds will still do well, as they have proprietary data that is hard to replicate. The traditional financial markets are also still constrained by regulation and manual processes.

But I believe that eventually, these institutions will use AI to replace the majority of their employees, including portfolio managers.

It won't happen overnight, but it will happen sooner or later.

My assessment at the time was that I had roughly a 4–5 year window. Once the foundational AI companies absorb enough industry talent, it will be challenging for new startups to enter this space. In some markets, such as the U.S. stock market, this trend is already very apparent. The level of efficiency a few years down the line will be almost unimaginable.

Soon, there will no longer be room for a "second place" in this world. I could continue working for top-tier institutions, but I would rather make a move in a field where I still have an edge.

So, I resigned and went all in on entrepreneurship. Later, that company became OpenForage.

Now, the window is rapidly narrowing. The pace of change is no longer gradual. What used to take months to progress now only takes weeks.

I don't believe that jobs will disappear entirely in the next few years. Humans still need humans. We are social creatures, and currently, humans still do not trust AI. Authority validation still needs to come from humans.

In the coming years, we may even see AI CEOs, but it will likely still require a human CEO to approve AI decisions. This "human validation" will trickle down through the organizational structure. Human managers will oversee a group of AI agents.

However, the logic of hiring will change. If the CEO finds it easier to command AI than to command you, then you are unlikely to be hired, and basic coding jobs will become increasingly hard to come by.

If you want to make yourself irreplaceable, you need to achieve two things. First, outlast AI on a timescale. For example, long-term strategic planning, complex decision-making, multi-year cycle management. Second, outscale AI in a systemic scope. The context of AI is still limited; they know many facts but struggle to understand the ripple effects of complex systems.

If you can think long-term, absorb information quickly, make strategic decisions, and collaborate effectively, then in the foreseeable future, you will still have a job.

The turning point is actually visible before it arrives. However, most people either do not look, see but do not act, or only react when the signals become deafening. By then, opportunities are often already priced in by the market.

Do not ignore the shifting ground, do not linger in a position that is losing advantage, all the while telling yourself to wait for a better time to act. The real opportunity rarely gives advance notice. By the time everyone is aware, the window is often already closed.

I saw the signal, I made the bet. Now, I am living in the outcome of that bet — for better or for worse.

[Original Article Link]