This September, venture capital firm a16z announced the launch of a new school: The Horowitz Andreessen Academy.
It is not trying to replicate a traditional university.
Its first cohort will not earn degrees or academic credit. Grades and test scores may still be considered, but what matters much more is proof of work: What have you actually built, studied, organized, created, or pursued?
That could be a piece of software, a research project, a community, an event, a work of art, or simply a difficult question that you kept following.
What students leave with is not just a credential, but a portfolio that shows what they have actually done, understood, and created.
Behind this model is a more fundamental question:
If AI can already help us access, organize, and even apply vast amounts of knowledge, how much of education should still be centered on simply knowing more?
The Horowitz Andreessen Academy offers a direct answer: let students do things.
Students spend much of their time pursuing projects, sharing their progress, receiving feedback, and deciding what to do next. They may start a company, build a robot, conduct independent research, write a book, create art, enter a new field, or spend time exploring a question that does not yet have an answer.
AI is not treated as something that should be kept outside the learning process. Instead, it is treated as a tool that can accelerate exploration and execution.
But there is a deeper issue here.
Human knowledge was not always organized the way it is today.
From Leonardo da Vinci to the early naturalists, the study of nature, engineering, art, anatomy, geography, and philosophy was often not divided into separate activities. The same person might sketch plants, dissect bodies, design machines, and observe water, birds, and rocks.
As knowledge expanded, we began to specialize.
Specialization brought enormous progress. Physics, biology, chemistry, engineering, and medicine divided into increasingly precise disciplines. Each field developed its own language, methods, datasets, and professional culture.
That specialization is one of the reasons modern science became so powerful.
But it also came at a cost.
We became increasingly good at going deeper, while gradually losing some of our ability to move sideways.
Today, someone may understand a particular protein, algorithm, or ecosystem in extraordinary detail, yet remain unaware that a neighboring field already contains a useful idea, method, or pattern.
AI makes this situation even more interesting.
Today’s large AI models are trained on the body of knowledge humanity has produced over the past several centuries. Much of that knowledge was created within highly specialized and increasingly fragmented disciplines.
As a result, AI can now help us enter the depth of many fields much more quickly.
It can explain materials science, write code, summarize ecological literature, and help us navigate history, mathematics, or engineering.
The harder question may no longer be simply:
How do we go deeper into one field?
It may increasingly become:
Which depths should we connect?
This is one of the questions that interests us most at SOS.
We are not trying to return to an era when one person was expected to know everything. Nor are we arguing against specialization.
Quite the opposite.
Because of AI, we may now have a new opportunity to combine two things that have become increasingly difficult to hold together:
the depth of specialized knowledge and the breadth of the naturalist.
AI can help us enter different bodies of knowledge quickly. Human effort can then focus more on observing, questioning, comparing, connecting, and judging.
To understand a forest, we may need ecology, sensing technology, imaging, history, and local knowledge at the same time.
To turn a biological structure into an engineering idea may require a natural historian, a materials scientist, and a designer to notice the same pattern from different perspectives.
The question is no longer only:
“Which discipline do I belong to?”
It may instead become:
“What other knowledge might be connected to what I am seeing?”
This is also the direction we continue to explore through SOS Education, Nature Salon, and our research programs.
Rather than placing students inside a fixed discipline first and then telling them which problems to solve, we begin with observation, curiosity, and real-world questions.
Students move across fields, work with people from different backgrounds, and eventually turn what they learn into something that exists in the world: a study, an artwork, a dataset, a tool, or another form of meaningful contribution.
In some ways, we are trying to recover the mindset of the naturalist.
But this time, we do not have to give up the depth that modern science has spent centuries building.
We can let AI help us enter that depth, and let humans reconnect it.
The Horowitz Andreessen Academy applies this model to entrepreneurship and technology.
At SOS, we are interested in extending it toward nature, science, and a broader culture of discovery.
The goals are not the same.
But both point toward a question that is becoming increasingly important:
As machines become better at navigating what humanity already knows, how should we educate the next generation to discover what has not yet been connected, and perhaps what has not yet even been asked?
Perhaps the most important educational skill in the age of AI is not simply the ability to learn.
It is the ability to explore.
