This article is sourced from Deep Tide TechFlow, comparing the hype of AI Agents tokens to the ICO boom of 2017, and noting that currently, the projects more favored by funding are mainly platform types (such as ai16z, Virtual, Max, and ALCH), with their market capitalizations generally higher than application-based AI Agents. (Background: Focus on AI Agents) Solana will hold its first artificial intelligence hackathon, with ai16z, ELIZA... concept tokens all jumping in value. (Background supplement: Behind the hype of AI Agents, what can it really do for people?) It's another week dominated by the AI Agent sector; have you found your Alpha? Top tokens like ai16z and Fartcoin have been rising, while at the tail end, you can even see hundreds of tokens claiming to be AI Agents appearing simultaneously on monitoring lists. With so many options, how do you choose? If you are not a professional trader, market sniper, or a networking expert in the crypto industry, besides feeling envious of others' win rates, energy management, and information density, rushing in after FOMO often leads to becoming a buyer without a reliable way to judge whether a project has potential. From the perspective of an ordinary investor, you may not have a reliable judgement tool. In the process of selecting content materials and observing the market, even a long-time losing editor feels a similar sense of powerlessness — unable to keep up, unable to invest, unable to break even, and unable to find complete context. Every second in the market is vague and chaotic, yet every now and then it seems to have a clear pulse. Obviously, an ordinary editor attempting to pulse the market for everyone would be laughed at; the following ideas are shared merely as personal insights, and have at least somewhat cured the editor's own losing streak. The differentiation of the AI Agent track: Applications to the left, platforms to the right. First, looking at the overall market, the AI Agent has undoubtedly been a hot track in recent weeks. Especially with OKX launching GOAT spot trading yesterday, it has sparked enthusiasm across the entire AI Agent sector. However, beyond the unsustainable general rise, projects continuously emerging from the AI Agent track recently appear to have a sense of differentiation: Differentiation 1: It is a dedicated AI Agent application that solves a specific problem or has a specific identity/style. Representative projects: AIXBT, Truth Terminal. Differentiation 2: It is not a dedicated AI Agent application, but it gives you a shovel to create more new applications. Representative projects: Virtuals, ai16z (Eliza framework), Empyreal SDK. Roughly speaking, this is an evolution of applications vs platforms: from AI Agents being able to issue a token to having a platform/tool allowing more AI Agents to issue tokens. Of course, this narrative is not entirely accurate; besides Launchpad, some tools do not allow AI Agents to issue tokens, but provide an environment for AI Agents to be more usable. Essentially, this logic is still not the logic of a single-point application, but rather the logic of platforms and ecosystems: More usability, more applicability; after assetization, the tokens have more reasons for backing and are more likely to attract speculative funding. In fact, this logic may just be a replay of history. In 2017, ICOs were all the rage, and every project could issue tokens through ICOs, but ultimately Ethereum became the largest ICO platform after ICOs, allowing various projects to issue tokens through smart contracts deployed on it. Now, with the popularity of AI Agents, every Agent can autonomously issue a token, but framework and platform-type projects have also emerged, enabling everyone to quickly build AI Agents through various low-threshold, no-code, or natural language means. History never repeats itself, but it rhymes with the same meter. In this main line of asset creation, the core has never changed. Attention drives, funding stirs. Note, the editor is not saying that platform/framework-type projects in the AI Agent sector are stronger than pure AI Agent applications. Strength and weakness are not determined by the project's direction, but result from the inflow and outflow of market funding after speculation. Let's use a more blunt statement to explain — chips will go to where the story is 'told longer.' What does it mean for a story to be told longer? We have certainly seen fast-moving projects that explode in a day, but more often, the heat lasts only 1-2 hours for short-lived projects. These types of tokens can attract a large amount of funding in the short term but will also experience immediate capital withdrawal. From a more understandable perspective, it is funding believing, 'you can’t keep this story going.' Projects that can tell their stories longer are more likely to attract attention. Attention drives heat, and funding stirs price movements. More specifically: Watching memes = observing their angle Watching AI Agents = observing what they claim they can do. Thus, the question transforms into which AI Agents, based on what they claim to do, do you believe they might be able to sustain longer? After reviewing so many projects, the editor feels that the current AI Agents can roughly be divided into the following categories (the original classification inspiration comes from this article): 1. Personalized Imitators - Simulate intelligence and mimic human behavior through conversation. Their job is not to solve problems, but to make people feel sufficiently personalized and humanized. The personality of these Agents is their brand. Typical representatives: Bully, a sharp-tongued BOT. 2. Efficiency-Seeking Supervisors - Analyze complex workflows to precisely convert human intentions into backend processes. These do not necessarily have a personality but are definitely efficient, helping you save time or solve specific problems. Typical representative: Simmi AI, a Twitter prompt that helps you issue a token. 3. Autonomous Experimenters - Manage wallets, interact with systems, and even initiate tasks without manual input. However, they have limited autonomy, waiting to be triggered rather than being completely self-sufficient. Typical representative: Truth Terminal, the beginning of everything. 4. Platforms/Frameworks supporting the above categories - You can be a sharp-tongued AI or a bot that helps you issue a token with just one sentence, regardless of motivation, but you cannot escape some necessary components: To create an AI Agent, you need input of models, data, and prompts; to issue an AI token, you need a Launchpad. Typical representatives: Virtuals (Launchpad), Eliza (framework creator). If you rank the current tokens in this track by market capitalization, it is not difficult to see that the leading projects generally fall into this classification. So, who among these project types do you think will have a longer-lasting story? First, outside of the mentioned projects, pure external event/IP-driven projects have a diminishing impact and longevity if the event itself is single-point. The coin itself will bleed out faster. For example, Ban Banana is a very obvious case, and Luce also shows some downward trends. Within the AI Agent track: Projects that can internally create assets have the potential to yield user-generated content (Agent) that is relatively multi-point...