news

---

  New Intelligence Report

The thing that AI entrepreneurs of this era worry about most is that their products will be raided by the "regular army"; the sentence they fear hearing the most is probably "This is a shell application."

Better startups are waiting to be acquired, such as。

Those who were not so lucky died on the eve of product launches by AI giants such as OpenAI, leaving behind only a life-saving motto to warn those who come after them:

Don't do what the giants will do sooner or later, otherwise, when they make their move, you will be crushed.

However, is this view, which is widely circulated in the entrepreneurial communities at home and abroad, really reliable? Where does this impression come from?

The AI ​​bubble is getting bigger and bigger, and the return on investment is slow to come. Is a technologically advanced model all we need?

If you also have these real questions, perhaps today's article will give you new ideas and the courage to do it.

Because a great company may also have started from a "shell". The importance of technology is self-evident, but the difficulty and importance of making a good product are often underestimated.

Most current products are just good enough and are far from being truly great.

In the entrepreneurial wave that has swept the world due to ChatGPT, "shelling GPT" is a great insult to an AI startup company.

"Shell" refers to products that have almost no independent technology and rely on the technology of others.

Three months ago, on the 20VC podcast, Altman said that any startup or product that tried to build within OpenAI’s blast radius would be crushed.

An important reason why Ultraman is so confident is that many current AI startups are built on the "basic model" provided by OpenAI. Of course, there are also basic models of other giants, such as Anthropic's Claude and Meta's Llama.

As soon as a large company’s model is updated, the small company’s products will be abandoned by users.

In fact, Altman is not the only one who holds this view. There are many voices that are pessimistic about AI start-ups.

For example, technology media reporters -

“Most AI startups are doomed to fail”

And executives at big tech companies --

And many enthusiastic Reddit netizens——

99% of AI startups only have prompt words

"Shell" companies do exist, and a considerable number of them have been crushed by the iterations of OpenAI models.

For example, Jasper AI builds AI copywriting tools based on the OpenAI model.

Before ChatGPT sparked the craze, their tool was highly praised and quickly reached a billion dollar valuation.

However, OpenAI updated ChatGPT last year to allow users to upload documents in various formats.

Once most users realized they could get the exact same functionality directly from the source, Jasper's revenue and valuation began to decline.

The Information counts the funding situation of the GenAI industry, among which more than 100 companies are building their own AI models, and 68 companies use OpenAI's models.

The logic behind this view is simple and intuitive - "shelling" cannot allow a company to establish product differentiation and "moat", and it is easy to copy and replace, and be washed ashore by the following waves.

But Eric Olson doesn't think so.

He published a signed commentary in Fortune, openly challenging Altman, "Sam Altman warned that OpenAI would 'crush' AI startups. I happen to run such a company, and I'm not worried at all."

Eric Olson is the co-founder and CEO of AI startup Consensus.

The goal of Consensus is to make it easier for people to access and search peer-reviewed scientific research. In layman's terms, it is equivalent to a Google Scholar + ChatGPT.

So far, Consensus has avoided several attacks from AI giants. But with OpenAI's entry into the search field with SearchGPT, some people think that the company's good days are over.

But Olson firmly believes that "the future of AI startups is bright."

Those doomsday headlines about startups, like those catchy headlines in other industries, are mostly just for attention.

Olson pointed out that the "shelling" issue should not be over-blamed.

Building products and companies with third-party technology as the core is not a bad thing in itself, and founders do not have to worry about becoming a "shell" in the early stages of the product.

In fact, it’s sometimes necessary to “shell” in the early stages in order to succeed. As a startup, your job is to make the “shell” thicker over time through design, user interface, new features, services, branding, etc.

This is not new in any field, and this development trajectory existed before the GenAI craze.

If we use today’s standards for AI startups to examine the giants that have risen before, they can also be called “shells” of various third-party technologies when they were first established:

- Salesforce is a "shell" based on Oracle database

- Box is a "shell" based on AWS

- Zoom is a "shell" based on Mac and PC cameras

- Delta is a "shell" based on Boeing aircraft

……

A new feature released by OpenAI doesn’t necessarily destroy you, because it’s also a new feature for you.

Your job as a startup is to add enough additional value around that feature to make it attractive and useful to users so that they pay for it over time. As the technology you rely on improves, so will your product.

Most things are just "shells" at first. There is nothing wrong with that. The only possible sin is that it is just a "shell" from the beginning to the end.

Nowadays, a large number of very eye-catching AI product demos have emerged on the Internet.

However, when these products are handed over to users, only a handful of them can truly satisfy them and solve problems.

This is because current AI products pursue "passing" or "good enough". As long as this is met, combined with a set demonstration environment, outstanding results can be achieved.

Before the advent of LLM, the core work of building excellent software products included hundreds of factors, such as a deep understanding of customers, elegant design with aesthetics, and thousands of lines of code for every edge case.

However, the emergence of LLM reduces the cost of product marginal intelligence to almost zero.

You just build a simple UI, add a feature or two, then add some OpenAI API calls and you have an amazing-looking product, built at a rate never before seen in human history.

However, from a founder’s perspective, even with LLM being so powerful, building a great product is still very difficult.

Just because it's easy to build software that looks good doesn't mean it's easy to build really good software.

In his entrepreneurial experience published on Zhihu yesterday, the great Li Mu also mentioned the difficulty of product building. Successful products such as Perplexity and Character.ai are still exploring business models.

It is a long process from technology to product, and it is normal to take 2-3 years. If we take into account the emergence of user needs, it may take longer. We focus on the present and explore the way in the fog, and remain optimistic about the future.

A more well-known example is Google’s much-criticized “AI Overview” feature, which forms a stark contrast with the rapidly rising AI search product Perplexity.

By some definitions, Perplexity is not a product with a moat. It only provides users with LLM interaction with search results.

In a world full of big models, couldn't Google, as the most powerful search engine in history and with its own flagship product, the Gemini series, add an LLM summary to the search results to put an end to Perplexity?

They can certainly try, and they do try. But so far, those efforts have been unsuccessful.

So, how did Perplexity defeat Google, which has core technology, by "shelling"?

The key point to understand here is that software products are not just about the top-level implementation. They are a collection of details that determine how they solve the user's problem.

Perplexity has mastered the details: the user interface has personality, but is extremely simple.

When you enter the search page, the cursor is immediately placed in the search box. The response time is almost instantaneous, and it even comes with a pleasant loading screen.

Google’s AI Overview lacks Perplexity’s obsession with detail. As a result, they haven’t won the same love from users.

This is the difference between "passing" and "excellence" - they may look the same from a distance; but when you zoom in, the two are very different.

For some niche demands, since there is not a large enough market, the technology giants will not bother to fully address them.

This creates space for startups to come in, innovate, succeed, and then eventually scale.

The creation of Consensus is based on this premise.

Google Scholar is the most widely used academic search tool in the world, but few people like to use it.

The reason is simple: it was a fringe product from Google that severely lacked attention and support from the company.

In contrast, a startup like Consensus can devote much more attention and effort to this, while Google has a million more important things on its plate.

Propose a solution to a niche problem that can stand the test of time. Even in the AI ​​era, this product idea is not outdated.

Even if we have to build products in the shadow of OpenAI's technology, this is a valid approach.

If the only thing that mattered was raw technical capability, then startups would never be able to succeed against giants with deep pockets and cutting-edge technology.

But the opposite has happened countless times. What really matters is the subtleties of the product, and you need to make users feel that you can provide professional solutions to their problems.

As Nat Friedman, a well-known AI investor (and Consensus investor) and former GitHub CEO, recently said at X:

“People hire a cleaning service to clean their office rather than a generic labor service, even though it’s essentially the same thing.” — Advice for AI startups

If you measure only raw ability, the average person and the employee of a cleaning service are virtually identical, with the only difference being the packaging—cheap materials (cleaning supplies), a bit of expertise, and the trust that this person has solved your exact problem before.

This difference will drive 99 out of 100 people to choose to pay extra for the cleaning service company's services.

People want to use things that are designed for specific things. This is probably the most inspiring statement an AI company founder can hear today.

Taking a step back, the three points mentioned above are not new advice. For decades, countless startups have relied on these core principles to succeed in the long shadow of existing enterprise technology.

The fear of big companies cornering your startup is a feature of startups, not a bug, and it’s for this reason, along with countless other reasons, that startup success is so hard.

We all now have an exciting new technology in our possession, and the natural reaction is to believe that “everything is about to change.”

But the reality is that some things will change, while most things will remain similar to the past.

As before, there is room for creating exceptional companies and products alongside the eye-catching giants, and the so-called "blast radius" that Altman claimed is actually smaller than most people think today.