Anxiety of AI gaming unicorns: AI is evolving too rapidly, challenging traditional VC decision-making
Yesterday, I had a long conversation with a friend who had just secured a very large - scale financing and is working on AI games. That kind of conversation wasn't a high - density technical exchange. It was more like the shared resonance of two people deeply entrenched in the industry, responding to the rhythm of the same era.
The most intense feeling I had was actually just one: The development speed of AI has begun to significantly exceed the cognitive boundaries of us practitioners.
01
If we rewind more than a year ago, AI was still in a rhythm where there was a "blockbuster every few months". After a model release or a product breakthrough, we would collectively marvel at how fast the technology was progressing and then return to our regular work rhythms.
However, starting from 2026, this rhythm was completely shattered.
Now, it's more like this: Almost every week, there is an update that can change the product form, the way we work, and even the cognitive framework. It's not about small features or parameter optimization, but a real leap in the ability to reconstruct the production process.
In the past week or two, I've been following these changes in an almost unhealthy way. I sleep about four hours a day on average. Apart from the necessary work and daily life matters, I spend a lot of time reading about the latest model developments, product releases, and research trends. Even so, I can only barely tell myself that I've probably kept up. But strictly speaking, I've actually started to fall behind.
What really made me realize the strong impact of these changes on the real world was the state of my friend.
He now manages a team of nearly forty people with a very clear direction, which is to develop AI - driven game products. Judging from the financing scale, staff allocation, and organizational maturity, this is already a quite healthy startup. AI has indeed significantly improved their production efficiency. Many modules that originally required the collaboration of four or five people can now be delivered by a single engineer in a very short time.
On the surface, this means that the team no longer needs to expand its R & D scale.
But the reality is just the opposite. His sense of anxiety is stronger than mine.
The reason is simple. The update rhythm of AI has become so fast that every leap in model capabilities directly affects the product design approach, system architecture, gameplay design, and even the business assumptions themselves. More importantly, he is a manager, not a front - line researcher. He doesn't have enough time and energy to systematically track every technical route and the capability boundaries of every new model.
So he has started to seriously consider a configuration that sounded extremely luxurious and even a bit "detached from the business" a few years ago - specifically recruiting several people who are only responsible for following up on the cutting - edge AI research, rather than directly participating in product development.
Their only task is to ensure that the company doesn't get left behind by the times in terms of cognition.
This really struck a chord with me. Because it means that in the stage of rapid AI evolution, "the ability to update cognition" itself is becoming a scarce resource.
02
During our conversation, we also talked about the overall landscape of domestic AI companies and very naturally reached a highly consistent judgment: Among Chinese AI enterprises, those most likely to enter the first echelon first and even lead in some areas will, with a high probability, come from the ByteDance ecosystem.
This judgment doesn't come from a comparison of certain model parameters but is more of a realistic and organizational - level judgment.
First, the overall execution efficiency and engineering capabilities of the team are extremely strong, operating on a different efficiency level from many traditional large - scale enterprises.
Second, the decision - making level has a very strong willingness to invest in new technologies and a strong sense of enterprise.
Third, and perhaps the most easily underestimated but extremely crucial point in the AI era, is the scale and diversity of the data it holds.
More importantly, it's the product route.
While many teams still focus highly on the single goal of "the intellectual ceiling of the model", ByteDance's core products have obviously chosen a route that is more oriented towards mass - use scenarios, emphasizing popularity and real - world usage frequency, rather than simply pursuing a leading position on the capability list.
From the perspective of user scale and activity, this route has begun to show very obvious advantages. At least in the domestic market, the penetration rate of Doubao's AI products for ordinary users is likely to be in an absolutely leading position. In fact, I personally think that in terms of the scale of active users, it may have even surpassed ChatGPT.
From a business perspective, we further discussed a more fundamental question: So far, has AI really created a new business model?
Our conclusion is quite conservative.
To date, AI itself has not brought about a new business structure beyond the Internet paradigm. What it has brought more is a complete reconstruction of the existing business system.
Especially in the Internet industry, this reconstruction seems particularly reasonable. Traditional businesses generally suffer from complex processes, fragmented systems, and redundant organizations, and AI provides a very direct and scalable means of cost - reduction and efficiency - improvement.
In this context, those who can truly maximize the benefits of AI are not the newly - established companies from scratch, but the technology giants that already have mature business models, stable user entry points, and complete ecological systems.
This is why it makes perfect business sense for Alibaba to deeply integrate Tongyi Qianwen into its business system or for Google to incorporate Gemini into its product ecosystem. They don't need to build a new business model; they just need to use AI to significantly improve the efficiency and competitiveness of their existing systems.
For these companies, AI is more like a "super - efficiency engine" rather than a model revolution.
03
However, from the perspective of entrepreneurs, AI still presents an unprecedented opportunity.
The reason is also very straightforward. The productivity structure has undergone a fundamental change.
What used to require a hundred - person team can now be accomplished by ten people. Looking further back, it's very likely that a single person can build a complete product prototype, system architecture, and even an early - stage business closed - loop. This means that the threshold for innovation is being significantly lowered, the cost of trial - and - error is dropping sharply, and the product iteration speed is increasing exponentially.
But behind this great liberation of productivity, there is a problem that is very easy to overlook or even deliberately avoid: the number of good projects will explode.
In the past, there weren't many truly worthy startup projects in the world that deserved serious attention every day. Resources, technology, and organizational capabilities themselves formed a natural filter. But with the help of AI, this filter is quickly becoming ineffective. Dozens, hundreds, or even thousands of highly - developed product prototypes may appear every day.
This will have a huge impact on the traditional investment and financing system.
I can hardly imagine how the traditional VC decision - making model, which relies on manual judgment, long - term due diligence, and in - depth support, can maintain its original efficiency and return structure when the number of high - quality projects doubles.
Further extrapolating, we even began to envision a more radical scenario: When almost everyone can build their own products, AI products themselves may gradually evolve into a new form of attention - economy market, similar to today's Meme ecosystem.
But there is a key difference between the two. Future AI products will be truly usable and may complete a business closed - loop in a very short time. That is to say, an AI product may accumulate the necessary revenue for its entire lifecycle within a few weeks.
In this highly short - cycle and high - concurrency startup model, financing and exit will instead become the biggest frictional costs. For this reason, we even seriously discussed a previously very radical judgment: In the future, the investment and financing process of some AI projects is likely to be completed more through the Crypto system.
The reason is not complicated. From a purely economic efficiency perspective, it may be the infrastructure that is currently closest to the ultimate efficiency of capital flow that we can see.
After discussing the industrial structure and startup model, we returned to a more fundamental question: How far are today's large - scale models from true AGI?
Our judgment is very straightforward. In terms of "intelligence" itself, the current mainstream models are already very close to our intuitive expectations of general intelligence.
What may really trigger the next qualitative change doesn't entirely come from the scale of the model itself but is more likely to come from two directions.
First, it's about permissions and long - term memory.
When an AI has extremely high system permissions, can take over personal devices, account systems, and behavioral data, and has the ability of long - term memory, being able to continuously learn and understand a person's preferences, habits, and decision - making patterns, its capabilities will no longer be just a "question - answering tool" but will transform into a true personal intelligent hub.
Second, it's about the access to high - quality, real - time, and highly professional data sources.
I have very specific personal experience with this.
About a month ago, when the movie "Avatar" was highly discussed, I asked several mainstream models (Doubao, ChatGPT, Grok, Gemini) about the latest domestic and international box - office data and market performance. The results were very interesting. Only Doubao provided very accurate and up - to - date data. Most of the other models just grabbed some obviously outdated news reports based on general Google search results.
The root cause doesn't lie in the model's capabilities themselves but in the data source structure. General search can only provide publicly available information with low - frequency updates, while truly high - quality data comes from highly professional data service systems.
The significance behind this is actually very important. In the future, the in - depth cooperation between large - model companies and professional data companies is likely to form an extremely crucial but not easily noticed moat. When users continuously receive accurate, real - time, and reliable data feedback on highly professional issues, the perceived level of intelligence will significantly leap, even if there is no revolutionary change in the model itself.
From the perspective of the user experience, this leap is even more intuitive than a simple improvement in reasoning ability.
04
Of course, we also talked about many more detailed things. Some of them involve the specific strategies and internal layouts of his company, so I'm not suitable to elaborate. But there is one conclusion that we almost completely agree on.
The next five to ten years are very likely to be the period with the greatest magnitude of change in human history.
It's not just a change in some local industries but a systematic reconstruction of the social operation mode, production structure, and personal ability structure.
We're both very anxious. But at the same time, we also feel very lucky.
Being in such a rapidly changing era is itself a rare opportunity.
What's really dangerous is not failure but the illusion of stability.
I'm becoming more and more convinced that those who believe their career paths, ability structures, and industry positions are solid will probably pay a price for this judgment in the future.
In this era, I prefer to position myself as a "speculator".
Not speculating on assets but speculating on directions.
When the trend changes, I can quickly adjust my cognitive structure; when the underlying logic shifts, I'm willing to abandon the existing path and rebuild myself.
It's not about betting on one - time success but about ensuring that I'm always in the midst of change.
This article is from the WeChat official account "Alpha Workshop Research Institute". Author: wsjack. Republished by 36Kr with authorization.