Author: Wang Jianshuo, Founder of Baixing.com
Many people blurt out phrases like 'large models are everything.' I'm not so convinced.
Every time I hear phrases like 'devour everything,' I think it's mostly because our understanding of the future hasn't reached that level, leading us to make such broad statements. Otherwise, how could one thing devour everything? Take the internet for example—it has been touted as devouring everything for so many years, has it truly devoured everything now? So which is it, the internet devouring everything or large models devouring everything? Both are devouring, leaving nothing left?
So, I'd rather put it differently: it is a very important foundational layer. Without this foundation, the entire world cannot develop, just as the internet cannot exist without the underlying backbone network, or using electricity is impossible without power plants. This I acknowledge.
But once the foundation is in place, the real excitement happens on top of it.
Take electricity. Once electricity was generated, what were the first applications people saw? The light bulb. Thomas Edison lit the first one, and then it just kept shining, shining, shining. If the world ended there, with just one light bulb, I could totally say: the power plant is the core of the entire world, the power plant devours everything.
But that's not how it happened. Later came motors to drive machines; then you realize that once you have a fundamental thing like electricity, countless electrical appliances will grow on top of it to use it. Washing machines are for washing clothes, televisions are for watching TV, vacuum cleaners are for vacuuming—they are all applications of electricity. Without electricity, none of these things would exist. But if you say 'electricity devours everything,' I don't believe it.
Large language models are the same. They provide foundational intelligence. But this intelligence must be placed into a specific, scenario-oriented 'machine' or 'appliance' to truly function and change the world.
Claude Code is for writing code, Claude Design is for design, VoiceDrop is for writing articles. The same large model, placed into different 'appliances,' solves completely different problems.
Just having electricity, just having water, without a washing machine, clothes still can't be washed. Imagine, the power plant produces a massive amount of electricity, the electricity is incredibly powerful, and then what? Without a washing machine, can this pile of electricity wash clothes by itself?
Intelligence is great, but most things in the world require the combination of multiple elements to work, just as a washing machine needs to combine elements like electricity, water, and even a drum. Large models might replace a lot in the field of software, but there aren't many application scenarios in the world that require only one element.
Let's take a current example. We have large models now, but just having a large model isn't enough. There needs to be a layer on top of it called Harness—this layer has only recently emerged—which interacts with code to finally form something truly usable. A large model by itself cannot write code. Of course, the core of something like Claude Code, honestly, I could write it in fifty-odd lines, add a few more for longer versions, and it could run to write programs. But you must see: with only the large model and without this outer layer, it's still not user-friendly—meaning, the intelligence of the large model, if not combined with the code execution capabilities provided by an operating system, relying on the large model for calculations is uneconomical and sometimes even impossible.
The core value of this interface layer is to help us put that intelligence, akin to electricity and water, into a specific application scenario, turning it into a machine that solves a concrete problem.
Having said that, of course, I don't completely disbelieve the logic behind 'devour everything.'
What it mainly refers to is existing software. Up until now, we have piled up a very large layer of software—things pieced together with many rules, forms, buttons, workflows. A lot of filters, fixed templates, backend operations, many SaaS detection functions. And the various things we used to know as 'M' systems, whether CRM, HIS (hospital information system), all sorts of things called 'systems,' 'software,' or whatever, a whole lot.
This layer, I believe large language models will indeed devour quite a bit of.
Why? Because these software were originally formed by solidifying and repeatedly executing clear, computer-executable instructions—that's what we call software. And this is precisely what large language models are most adept at tackling.
But.
Beyond software, within this layer, there are many other things. Customer information. Execution capabilities—like when you book an airline ticket, the actual capability to move a plane, to transport people from here to there. And trust. Many things in the physical world. These, I don't think will be devoured.
After devouring that layer, it actually opens up an even larger space—on top of it lies the new type of software.
The new software will definitely have a fluid interface; it likely won't solidify as many rules as before. After handing all those rules to AI, think about it, before we could achieve something like Salesforce's CRM, which was already the pinnacle of human effort, requiring immense struggle. If this part becomes relatively easier to solve, then what everyone will work on next is unlocking even more imagination, more possibilities on top of that—and that part is precisely what we haven't seen yet.
This is the mistake we often make. When a new technology arrives, because we can't see the bigger path that follows, we can only focus on the immediate part. We see the leaf but not the forest.
Forget such trend judgments. I still remember in 2004, a group of friends were complaining together, saying that the entire internet could never produce companies larger than Sina, Sohu, and NetEase, that the internet was almost over, that they would monopolize everything. And look how many years have passed since then? The world turned upside down. We would be shocked by our own shortsightedness back then.
So my view is this: Are large models important? Yes, they are a foundational layer, the main focus of recent times. But once they become stable, continuously provided resources, they will require various 'machines' and 'appliances' on top to solve specific problems. That thick layer—where it's used, how it's used—will be the mainstream of the second wave of this trend.
The phrase 'devour everything' is far too imprecise. What thing, what social form, what technology in the world has ever truly devoured everything?
Finding the opportunities within the areas it does devour—that is what truly matters.
