"Have you raised a lobster yet?" Lately, when Web3ers greet each other, this phrase is probably used eight or nine times out of ten.
At the beginning of 2026, since robots stunned the audience at the Chinese New Year Gala, a new generation of AI Agents represented by OpenClaw has become the new toy for tech enthusiasts. Some use AI for customer service, some use AI to write code, and some have even started trying to use Agents to simulate an entire set of "digital employees." A concept frequently mentioned recently on various internet platforms is the "one-person company"—where one person, through an AI workflow, can manage work that previously required a small team.
The Web3 space, of course, hasn't been idle either. Lately, if you look at industry media, you'll find many projects are also starting to focus on AI Agent narratives. Some are researching how Agents can directly call on-chain assets or contracts, some are working on payment, identity, or financial infrastructure for Agents, some are discussing an "Agent economic system" where AI can participate in the network like users, and some have started shouting the new slogan of "Web4.0."
Reading this, it actually feels very familiar.
They say the fashion industry is cyclical, but who would have thought the tech world (or the crypto world) would be the same? Remember during the bear market that started in 2022, when ChatGPT exploded overnight, AI suddenly became the topic everyone was talking about. The Web3 circle, of course, didn't stay idle either; soon, a bunch of new concepts emerged, like AI Agents, AI traders, automated strategies, and so on. It seemed that just by touching on AI, you could tell a new story. But this excitement didn't last long. Later, when the crypto market started rising again, everyone's attention quickly returned to Crypto itself.
And this time, in the second half of 2025, the crypto market is showing bearish trends again, so Web3 is looking for new narratives to take over.
However, in Portal Labs' view, the problem lies precisely here. When a narrative starts to become popular, many Web3 startup teams aren't making technical and business judgments; they're making narrative judgments: whichever concept is hot, they do that. And then they stumble—
Many teams only realize when actually pushing their projects forward that while concepts can be built quickly, products are hard to land. Where are the users? What is the specific scenario? How to generate sustained revenue? Can investment be secured? These questions often only emerge after the project has been underway for some time.
By the time the hype fades, what's often left on the market is a field of projects that haven't found product-market fit. Some products remain in the demo stage, some barely launch but can't find users, and some simply disappear along with the narrative. In the short term, it looks like a new track has opened up, but looking back after a while, what truly remains is actually not much.
Because of this, whether to continue deep diving into Crypto or pivot to AI has become a dilemma. Choose the former, and the market isn't great; investment may not yield returns. Choose the latter, and there's no solid foundation. The technical barriers, talent structure, and competitive environment of AI are all different from Web3. The technical stacks, product experience, and community resources that many teams have accumulated over the past few years are actually built within the Crypto system. A complete pivot to AI would mean re-entering a completely unfamiliar field. From model capabilities, data resources, to engineering teams, almost everything needs to be rebuilt.
More realistically, the AI赛道 itself is already very crowded. Whether it's large model companies, traditional internet enterprises, or a large number of startup teams, huge resources have been invested in this field. For a startup team originally in Web3, if they enter this market just because of a narrative shift, they can easily find themselves with neither technical advantages nor industry resources.
Actually, for many Web3 startup teams, there is another practical path. They don't necessarily have to转型做AI; they can continue on their own Web3 path while thinking about what capabilities Crypto can补上 in the AI system.
If you look closely at the current wave of AI development, you'll find that many key links haven't been fully resolved.
The most typical is data. Models are getting stronger, but where does the training data come from? Is the data trustworthy and compliant? Especially, how can AI Agents achieve 1v1 customization? These issues一直没有没有一个很好的机制. For AI that relies on large-scale data training, this is a fundamental problem that has long existed.
Another example is identity and collaboration. When AI Agents start participating in task execution, automated trading, and even operational decision-making, they themselves也需要身份、权限以及协作规则. Who can call a certain Agent? How do Agents分工? How to settle after executing tasks? These issues本质上都涉及到开放网络中的身份和价值分配.
There's also the payment problem. Once AI Agents start autonomously calling services, obtaining data, or executing tasks in the network, it means they need a micro-payment system that can settle automatically. In the traditional internet system, such a payment structure is actually very difficult to achieve.
These seem like AI problems, but many solutions actually already exist within the Crypto technical system. Whether it's data incentive networks, on-chain identity systems, or open payment networks, these are precisely the directions Web3 has been exploring for the past few years.
If Web3 startup teams really intend to try these directions, there are several things they must figure out first.
First, look at the team's own technical capabilities. Different Web3 projects have vastly different technical积累. Some teams are good at building on-chain protocols, some have long been working on data networks, and others are more focused on application-layer products. If a team has been working on data-related infrastructure for the past few years, such as data collection, data extraction, or data markets, then extending around the data layer of AI would be relatively natural—for example, data contribution networks, verifiable data sources, or incentive-based data markets for models. If the team was originally more focused on on-chain protocols or infrastructure, then they could consider working on the operating environment for AI Agents, such as on-chain identity for Agents, permission management, task execution protocols, or providing automatic settlement and payment capabilities for Agents. For teams that are already working on application-layer products, such as trading tools, content platforms, community products, or consumer applications, AI is more suitable as a capability layer embedded into the existing product system. For example, using AI to enhance data analysis capabilities, automate operational processes, or use Agents to complete functions that previously required manual handling.
Second, look at whether there is a real business scenario. Many AI projects disappear quickly not because the technology isn't good, but because there was no clear use case from the start. The concept can be talked up hotly, but where are the people who actually need this product, why do they want to use it, and why are they willing to pay for it? These questions are often not seriously answered. Some concepts are discussed a lot in the industry, like "AI+Web3," "Agent economic system," "AI trader"— they all sound grand, but if you dig one layer deeper, the number of stable user groups that actually exist is not large. On the contrary, some needs that seem less "sexy," like data processing, automated operations, information filtering, or task execution, have long existed in real business. Precisely because of this, when judging whether to enter a certain AI direction, rather than first looking at whether the concept is hot, it's better to first look at the scenario itself: Is this a long-standing business problem? Is someone already paying for it? And can AI truly improve efficiency in this环节? If these conditions are met, then this direction is more likely to go from narrative to product.
Going further, it is also necessary to see if the Web3 startup team has the resources to truly enter these areas.
The directions mentioned earlier—data, identity, payment—are本质上 not purely technical problems, but problems of network resources.
For example, a data network: if the team doesn't have a stable source of data, nor a user base that can continuously contribute data, then even if the technology is built, it's hard to form a real network effect. Similarly, if you want to build an identity system or collaboration network for AI Agents, you also need real developers, applications, or Agents to participate; otherwise, the protocol itself can hardly form an ecosystem. The payment and settlement system follows a similar logic. Once AI Agents start calling services, obtaining data, or executing tasks in the network, micro-payments will become very frequent. But such a payment network only makes sense when a large number of Agents and services coexist; otherwise, it remains just a technical module.
So for many Web3 teams, what really needs to be evaluated is not "is there technical space in this direction," but whether they can become a part of this network. Whether the team already has data sources, a developer ecosystem, or application scenarios often determines whether a project can truly enter the infrastructure layer of AI, rather than staying at the conceptual layer.
