DeepSeek Ignites AI’s “iPhone Moment” as Agent Tokens Integrate into Real-World Crypto
Overview:
In October 2024, Terminal of Truth launched $GOAT with a $50,000 Bitcoin donation from Marc Andreessen, sparking a surge in AI Agent token launches. Encryption technology gives Agents a borderless system for identity, assets, and settlements, operating 24/7 without human input. This drove the overall market capitalization of AI Agent tokens to tens of billions of dollars (currently down to around $9 billion), with both $ai16z and $VIRTUAL reaching market caps above $2 billion at their peaks.
During the ICO/token launch boom, leading Agent projects based on Launchpad generated a powerful flywheel effect, quickly attracting funds and users. However, this model struggles to retain real value, such as total value locked (TVL). As the initial ICO / token launch fades, user retention becomes challenging, leading to lower token demand, declining revenue, and falling token prices.
While industry leaders are exploring ways to integrate Agents with gaming, DeFi, and other use cases to drive their real-world applications, related products remain immature, and user growth has fallen short of expectations. Consequently, these projects cannot support crypto value through actual fee-based income. In other words, the current Agent ecosystem primarily attracts Degen players rather than genuine game gold farmers or DeFi miners.
Dividends on initial ICOs/token launches in the Agent sector have largely been exhausted, and capital is starting to exit with profits. Therefore, the market is in dire need of a new catalyst to break this deadlock, establish a new positive flywheel, and attract fresh capital and users. Against this backdrop, DeepSeek’s innovation brings new hope to the industry. Its groundbreaking technology enables large models to enhance their reasoning capabilities through pure reinforcement learning (RL), which significantly lowers costs and allows AI to potentially become an agent that can truly collaborate with humans. This progress could breathe new life into the Agent field and reshape the industry landscape.
This article will summarize and compare the different evaluation frameworks applicable to the Agent sector in the ICO/token launch and value maturity phases. It will also analyze how DeepSeek’s technological breakthroughs might have a far-reaching impact on the Agent sector.
1. Q4 2024: The ICO Era of Crypto + Agents, with a Focus on the “Existence” of Narratives, Teams, and Products
Since 2017, we have observed that blockchain projects typically evolve through ICO/token launch, post-launch, and value maturity phases. For example, AAVE raised funds under the name Ethlend in its ICO phase. It then went bankrupt, rebranded, and later had a resurgence during the DeFi summer. We believe the Agent boom will follow a similar trajectory.
During the ICO/token launch phase, project valuation largely depends on market expectations, narrative positioning, and community hype. Funds flow rapidly, with investors focusing on the project’s vision and the team’s execution rather than technical details. This phase is critical in determining whether the project can maintain its market appeal. Finally, in the value maturity phase, valuation is determined by long-term user growth, ecosystem scale, commercialization capabilities, and use cases of tokens. Fund flows stabilize, and the market’s focus shifts away from pure sentiment to the project’s sustainable growth potential.
Throughout the cycle, valuation undergoes drastic changes, from expectation-driven to product validation and eventually to long-term value maturity. Project teams must adjust their strategies in different phases to ensure they can continuously capture market value. In the post-launch phase, the market starts to pay attention to the projects’ actual implementation, technical progress, and ecosystem growth. Valuation gradually shifts from being hype-driven to being based on the products’ actual performance. Meanwhile, fund flows become more rational, and project teams’ communication capabilities and execution speed become critical.
This trend is especially evident in projects like Goat, AI16z, Virtual ecosystem, Swarm, Arc, Pippin, and DeFAI ecosystem. It highlights that in a fast-evolving market environment, a narrative-driven approach and execution capabilities often play a bigger role than the underlying technology in determining a project’s short-term valuation and ability to attract funds.
1.1 Narrative Positioning
In the AI Agent sector, the overall ecosystem can be roughly divided into four core segments: AI Agent frameworks, AI launchpads, AI Agent meme coins, and AI Agent applications. Each segment plays a unique role in the ecosystem’s development, working together to drive the ongoing evolution of the Agent field.
1.1.1 AI Agent Frameworks
As the core infrastructure of the agent ecosystem, an AI Agent framework offers developers a one-stop solution for building, training, and deploying agents through its modular architecture, distributed training network, and on-chain economic model. Its core value is reducing development barriers, accelerating AI adoption, and enabling ecosystem synergy. As the AI Agent sector gradually matures, the framework layer has become a critical point for capturing value. Specifically, the Eliza project creates a sustainable business model with its technological innovations and ecosystem design.
1.1.1.1 Modular Plugin Design
Eliza employs a modular plugin design, breaking its core system down into four key components: Adapter (data adapter), Character (agent personality), Client (message interaction), and Plugin (general functionality). This design lets the core runtime operate independently and allows developers to add plugins, characters, and adapters freely, thus supporting multi-chain compatibility (e.g., Ethereum, Solana, Ton) and multiple model providers (e.g., OpenAI, Llama). In multimedia generation, Eliza can create images, videos, and 3D models, support automatic generation of NFT series, and provide image description and analysis capabilities. Additionally, in infrastructure, it provides basic capabilities such as browser services, document processing, and speech-to-text.
Through this modular design, developers can easily customize their own plugins and tools, such as NFT generation tools or DeFi arbitrage strategies, and share 1%-3% of revenue via smart contracts. This not only gives developers flexible space for development but also enhances the ecosystem’s ability to capture value.
1.1.1.2 Technological Innovation
1.1.1.2.1 Internal Enhancements:
Eliza has achieved majorachieved major breakthroughs, especially in reasoning and decision-making. By introducing advanced technologies like Chain-of-Thoughts and Tree-of-Thoughts, Eliza’s AI agents can efficiently handle complex tasks and make more precise decisions. These technological breakthroughs have notably improved Eliza’s effectiveness in agent interactions, allowing it to handle increasingly complex use cases in the Web3 environment.
1.1.1.2.2 External Extensions:
Eliza also integrates technologies like RAG (Retrieval-Augmented Generation), vector databases, and web search to enhance the real-time information retrieval and processing capabilities of its AI agents. In the Web3 space, Eliza can access both on-chain and online data in real-time, keeping its agents accurate and timely.
1.1.1.3 Performance Assessment:
In the GAIA benchmark test (a platform specifically designed to evaluate the ability of AI agents to solve real-world problems), Eliza scored 19.42% in the test — below the top solutions — but this is impressive for a Web3-focused framework. Particularly in handling basic tasks (Level 1), Eliza achieved a completion rate of 32.21%, showcasing its solid foundational capabilities and technical maturity.
1.1.1.4 Ecosystem Influence and Market Acceptance:
From an ecosystem standpoint, Eliza has established a robust foundation, and several significant Web3 projects have built their AI agent systems based on Eliza. As of January 2025, the combined market capitalization of these partners surpassed $20 billion, reflecting the market’s strong recognition of, and genuine demand for, Eliza’s technology.
In addition, based on feedback from over 50 AI researchers and senior blockchain developers, Eliza outperforms other frameworks in the following key metrics:
- Support for model providers
- Chain compatibility
- Completeness of functionalities
- Integration with social media
1.1.2 AI Launchpads
AI launchpads operate like “hybrid incubators + trading platforms” in the crypto space, designed to lower the entry barriers for AI Agent projects and help accelerate their launch, financing, and promotion. With well-structured tokenomics and strong ecosystem synergies, these launchpads can generate a powerful positive flywheel effect.
Take the Virtuals Protocol as an example. Its protocol revenue has reached $60 million, showcasing the immense potential of the launchpad model. However, the success of launchpads is highly contingent on the continuous emergence of blockbuster projects. A prolonged lull without market hotspots can weaken market expectations and lead to valuation corrections. Therefore, during periods of weak market sentiment or sector-wide declines, investors should adopt a wait-and-see approach. Representative projects include VIRTUAL, CLANKER, VVAIFU, and MAX, which have played pivotal roles in driving the financing and market expansion of AI Agent projects.
1.1.3 AI Agent Meme
AI Agent meme coins thrive on culture and community support. While they don’t require much technical expertise, they can quickly attract funding and attention through trends and sentiment. By combining the dual narratives of AI and memes, AI Agent meme coins are highly speculative and have enormous potential to go viral, positioning them as key vehicles for rapid capital accumulation in the short term. Representative projects include BULLY from the Virtual ecosystem and TAOCAT and LLM from the Bittensor ecosystem. Fueled by community power and sentiment-driven speculation, these projects have experienced swift growth within a short time, showcasing the unique influence of meme culture in the crypto market.
Although meme coins lack long-term value support, they play an indispensable role in early-stage fund accumulation, brand communication, and community building of AI launchpads. At the same time, they can quickly gain traction during the initial surge of narrative hype, similar to Fartcoin in the Goat ecosystem. However, without new hot topics, the popularity of AI memes is hard to sustain over time.
1.1.4 AI Agent Applications
AI Agent applications focus on the implementation and commercialization of the technology in real-world scenarios, covering areas such as automated trading, asset management, market analysis, and cross-chain interaction. With their value primarily stemming from their user base and actual revenue potential, they are the most user-centric sector within the AI Agent narrative. Representative projects such as GRIFFAIN, NEUR, and BUZZ have already demonstrated immense potential in areas such as financial automation, intelligent decision-making, and DeFi protocol interaction. These applications not only improve the automation of on-chain operations but also provide users with smarter asset management and market analysis tools.
AI Agent applications lag behind launchpads in commercialization. They need more time to expand their user base and prove their revenue potential. In the future, as technology matures and user education advances, AI Agent applications are expected to become the most promising segment in the sector.
1.2 Commitment and Resilience Are Core Competitiveness for AI Agent Teams
In the AI Agent sector, teams like Virtual, SEKOIA, ai16z, and Swarm teams show how top developers drive project growth, tackle challenges, and build community trust. Even though some entrepreneurs may not come from the most prestigious backgrounds, they still deserve attention as long as they remain committed to this field, interact actively with the community, and frequently update product progress.
Before the emergence of the Goat project, the Virtual team had already dedicated itself to the AI Agent sector, accumulating substantial technical experience and honing its ecosystem-building capabilities through projects like Luna. This long-term technical accumulation and industry insight empowered them to quickly grasp market trends, continuously launch innovative products, and contribute to the development of the entire sector.
Anand Iyer, SEKOIA’s founder, also founded the renowned crypto VC Canonical. He has rich industry resources and a solid investment background. Canonical has invested in several high-profile crypto AI projects such as Gensyn, Ritual, Sahara, Nuffle Labs, and Tensorplex Labs, covering frontier areas like decentralized computing, AI infrastructure, and data privacy. Anand’s strategic foresight and extensive experience have provided robust support for SEKOIA in technological deployment and resource integration, helping it rise rapidly in the AI and crypto fields.
Despite facing community-wide FUD at one point, the a16z team has always maintained an open attitude and gradually turned around negative sentiment through worldwide developer incentives and community building. Leveraging their exceptional team capacity, they successfully attracted a large number of developers to their ecosystem, steadily driving product iterations and successful execution of market strategies.
The Swarm team has also weathered several rounds of FUD, but its founder has consistently engaged in proactive communication, stayed closely connected with the community, transparently disclosed project updates, and responded promptly to community concerns. To prove the project’s technical viability and ecosystem potential, the Swarm team rolled out the MCS ecosystem project to showcase the scalability and practical applications of its technology, ultimately gaining recognition and support from the community.
This trend continues in the subsequent DeFAI sector. Daniele, the founder of the top project Anon, is an iconic figure in the DeFi 2.0 era, having previously created projects with a market cap of over $2 billion, such as Spell and Wonderland. Similarly, Griffain was created by Tony, a seasoned Solana developer who had successfully launched blinkdotfun. The developers and teams behind these projects are still the key forces driving the sector forward.
1.3 Product Usability
In the Crypto+Agent sector, product usability is the key driver of rapid user adoption and widespread market penetration. Whether it’s the abstract layer of the DeFAI (Decentralized Finance + AI) field, automated trading agents, and AI-powered dApps or projects in the Agent sector like Eliza, GAME, Rig, and ZerePy, their success is inseparable from the meticulous refinement of product usability.
In the early stages of ICO/token launch, market demands for functionality and technical sophistication are not as stringent as in the maturity phase. The key lies in ensuring the usability of core functions and lowering technical barriers, thus attracting target users. At this stage, users care more about whether the product is “usable” rather than whether it is “perfect”. As long as it is somewhat functional, coupled with frequent iterations and rapid updates, initial user stickiness can be established, which in turn drives ecosystem expansion and value growth.
For example, Virtual, a representative of the metaverse and gaming sectors, has made it possible for non-technical users to quickly create and deploy AI agents through its no-code and low-code development environment, driving the rapid adoption of its ecosystem. Meanwhile, ai16z has lowered the barrier to developing complex AI agents through its massive developer community and powerful TypeScript tech stack, helping developers quickly build products that are competitive in the market.
In the DeFAI space, the usability requirements are even higher. Griffain has simplified on-chain interactions, allowing users to execute complex on-chain tasks effortlessly through an intuitive interface, making DeFi significantly easier to use. Almanak has optimized automated yield strategies, helping users maximize returns without the need for manual intervention. This showcases the immense potential of AI in financial automation. Mode Network has demonstrated the scalability of AI-powered finance automation, enabling the automated execution of DeFi transactions with over 129 AI agents. This validates the universal applicability and scalability of AI in the financial sector. The Hey Anon ecosystem integrates various automated yield optimization tools and personalized agents to adjust investment portfolios in real time based on market data, thus boosting efficiency. With strong usability, frequent iterations, and an open developer community, Hey Anon has attracted numerous users and developers and rapidly integrated with multiple traditional DeFi projects.
2. DeepSeek Proves That Large Models Can Improve AI-Human Collaboration Through Reinforcement Learning (RL), Directly Benefiting Agents
DeepSeek-R1’s release has brought about a revolutionary breakthrough in the AI field, proving for the first time that large models can make leapfrog progress in reasoning ability purely through reinforcement learning (RL) without relying on traditional supervised fine-tuning (SFT). This innovation has blazed new paths for the development of AI agents, significantly enhancing their intelligence and application potential.
The success of DeepSeek-R1 boils down to its unique training method: it forgoes the traditional supervised fine-tuning phase and optimizes the model directly through reinforcement learning. This method enables AI to go beyond merely mimicking the human mind to develop true independent reasoning and self-reflection capabilities.
- Traditional Method (SFT + RL): It’s like teaching a child math. First, you buy a workbook (SFT phase), letting them repeatedly practice solving problems with standard solutions. Then, you hire a tutor (RL phase) to correct their mistakes and adjust their problem-solving approach.
- DeepSeek-R1 Method (Pure RL): You send the child straight into the exam, with no preview or example problems. For every incorrect answer, the judge immediately points out the error (reward signal). The child learns through trial and error and independently summarizes the optimal problem-solving strategy.
This approach allows the model to develop its capabilities in leaps and bounds and acquire self-learning, insight, and efficient adaptation:
- Self-Learning and Insight: DeepSeek-R1’s ability to gain insights demonstrated during RL training proves that the model can engage in deep reflection and adjustment when faced with complex problems. This trait enables AI to provide more creative and profound solutions rather than mere task automation when collaborating with humans.
- Efficient Learning and Adaptability: Through technologies like GRPO (Group Relative Policy Optimization), DeepSeek-R1 can quickly adapt to new environments and tasks. This efficient learning mechanism endows it with robust adaptability in the constantly evolving crypto market, allowing it to respond to market fluctuations in real-time and optimize its decision strategies.
DeepSeek’s success signifies that the computing power threshold is not as high as previously imagined. After DeepSeek’s launch, several labs, including those at universities in Hong Kong, successfully replicated its results. This indicates that the cost and entry barriers for large models have been lowered, enabling crypto startups to join the competition in large AI models and develop large models tailored for the crypto industry.
The story of Crypto Agents may have just begun because if large models can iterate so quickly, then the current lead of AI agents is only temporary. If these agents fail to keep pace with the trends and technological iterations, they could easily be surpassed by newcomers with the latest technology and models.
In addition, Crypto Agent projects may choose to integrate LLM-based DeepSeek R1 and abandon the expensive, closed-source OpenAI models. This is because the Workflow system typically involves large amounts of token consumption and context information (averaging >= 10k tokens). For this reason, using high-priced OpenAI or Claude 3.5 models for Workflow execution is extremely expensive. Until Web3 users see real value capture, this premature overspending actually harms the product.
With DeepSeek-R1’s breakthrough in reinforcement learning (RL), the intelligence of Crypto Agents will be directly enhanced. The multi-task learning capability of DeepSeek-R1 enables these agents to handle more complex on-chain tasks, such as cross-chain asset management and DeFi protocol interactions. They are no longer limited to single-chain or simple transactions but can operate flexibly in multi-chain ecosystems and achieve broader synergies. Furthermore, RL-based AI models can better understand user needs and preferences, offering personalized asset allocation advice and investment portfolio optimization services. More than mere execution tools, agents have become users’ intelligent assistants, greatly enhancing user experience and trust.
3. 2025: Crypto+Agent Value Maturity Period, with Protocol Revenue and User Adoption as Key Factors
After the initial launchpad model was launched, more competitors followed suit. As a result, market funds were quickly diluted, leading to a decline in overall liquidity.
At the same time, most Agent projects struggle with inefficient customer acquisition, lack of real-world use cases, and insufficient continuous product innovation. As a result, they fail to establish sustainable business models and will eventually be phased out by the market. This situation has amplified the market divide over the future value of AI Agents. Investors and users remain cautious and are less confident in the long-term prospects of this sector.
As the Crypto+Agent sector enters its value maturity phase, simply relying on narratives and product usability no longer suffices to attract and retain users. Even if the founding team has a strong industry background, projects may still experience a crash after the initial FOMO-driven surge as the team cashes out or fails to sustain growth. Only those Agent projects with stable revenue streams and the ability to cover AI reasoning costs will survive and thrive in the competition.
At this stage, actual revenue and user data will become the primary indicators of a project’s value, rather than market sentiment or the team’s past success.
Directions that deserve future attention:
Multi-Agent Systems and DeFAI (Decentralized Finance + AI):
- These two directions highlight AI’s tremendous potential in complex decisions and financial automation, demonstrating possibilities for long-term growth.
Key Framework Functional Updates:
2. It is crucial to keep an eye on the technological iterations of core AI Agent frameworks like ELIZA and ARC. Upgrades to these frameworks may give rise to new use cases and business models.
New Infrastructure Projects:
3. In particular, AI models, data layers, and computational infrastructures designed specifically for crypto scenarios will become bedrocks supporting the next-gen AI Agent ecosystems, improving model efficiency and intelligence.
The AI Agent sector is shifting from the “narrative-driven” early phase to a mature phase where the value is validated through data and revenue. The survival of a project through varying market conditions hinges on its sustainable profitability, genuine user demands, and a commitment to continuous technological innovation.
3.1 Infrastructure 2.0: Models Place High Requirements on Teams, with Growing Demand for Data Layers
Lumo Labs has developed an advanced AI model tailored for the Solana ecosystem, aiming to help developers, investors, and project teams analyze market dynamics, assess risks, and devise strategic plans more efficiently. Deeply integrated with Solana’s core infrastructure, the model can process on-chain data in real-time and supports multidimensional, intelligent analysis of DeFi and other on-chain protocols.
Key features include:
- Market Trend Prediction: It can analyze historical data and real-time market dynamics to forecast future trends.
- Intelligent Asset Allocation: It can optimize investment portfolios based on risk preferences and market conditions.
- Cross-Chain Liquidity Analysis: It can monitor liquidity across multi-chain ecosystems to identify arbitrage and optimization opportunities.
- Risk Modeling: By combining historical data with real-time indicators, it constructs accurate risk assessment models, helping users make more informed decisions in the volatile crypto market.
However, Lumo’s model has yet to see widespread implementation in practical use cases, and the team has not disclosed their real identities. It is recommended to keep an eye on its progress, particularly to observe whether it is committed to developing models suited for crypto business scenarios with a focus on **reinforcement learning (RL)** training so as to uncover potential development opportunities.
At the same time, cookie.fun has built a vast AI Agent monitoring network to track over 1,000 active AI Agents and create a robust indexing and data layer. The platform aggregates on-chain and off-chain data from multiple sources. In addition, it utilizes advanced data analysis and machine learning models to evaluate the performance of each AI Agent across key metrics like Mindshare, Market Share, Intelligent Engagement, and On-chain Activity.
The key highlight of cookie.fun lies in its data aggregation and intelligent comparison engine. The platform collects real-time data from Solana and other blockchain ecosystems, including on-chain transactions, social media trends, developer activity, and fund flows, and then creates detailed analysis profiles for each AI Agent. These profiles not only reveal the strengths and weaknesses of agents in specific areas but also provide insights into their adaptability in varying market conditions.
In addition, cookie.fun supports a mechanism of competition and ranking among AI Agents, helping users quickly identify potential winners in the market. Through a visual dashboard, users can track the most promising agents in real-time and assess their performance in areas such as mindshare, activity, and market penetration. The platform also offers an intelligent recommendation system that suggests the most suitable AI Agents based on user interests and investment preferences. This system assists investors and developers in seizing opportunities in the rapidly evolving crypto ecosystem.
3.2 Framework Upgrades in Progress: The Ultimate Winning Hand Lies in Key Features that Attract Users
Having accumulated sufficient funds in the first wave of asset growth, frameworks now have enough capital to develop ecosystems and upgrade products. The key winning hand is whether they can introduce unique features to create new AI agents that attract users.
Arc, a Rust-based rig framework for AI Agents, is not entirely dependent on the Solana network. It recently became one of the projects supported by the Arbitrum Stylus Foundation and co-hosted a grant event with the Solana Foundation. It also participated in ETHDenver’s hackathon alongside Monad.
The Eliza framework is also continuously upgrading. The upcoming update will introduce the following new features:
- The introduction of modular architecture offers developers a more scalable design and supports the flexible integration of various functional modules, such as natural language processing, automated decision-making, and data analysis. This simplifies the development process and improves system composability.
- Eliza has enhanced its multi-chain compatibility, especially optimizing its integration with mainstream blockchains like Solana and Base. This enables AI Agents to operate seamlessly across various chains, fostering cross-chain interaction and asset management.
- In terms of smart contract integration, Eliza supports the automated execution of on-chain tasks, such as DeFi protocol interactions, asset management, and governance voting, to raise the intelligence of on-chain operations.
- Eliza incorporates TEE and ZK technologies to bolster data privacy and computational verifiability. The framework also supports the autonomous evolution of AI Agents, as well as real-time model fine-tuning and performance optimization, thereby boosting their adaptability to the dynamic market.
- As for developer tools, Eliza provides an upgraded SDK and API documentation, complemented by visual monitoring tools to help developers quickly get started and track Agent performance. The open plugin marketplace encourages the community to contribute functional plugins, thus expanding use cases and driving the diversified development of the Eliza ecosystem.
3.3 DeFAI Is Closest to the Trading Side, But User Acquisition Capacity Is Still Insufficient
Currently, DeFAI products focus on addressing the complexity of DeFi’s UI/UX through natural language interfaces. These products enhance users’ profitability, allowing them to earn more through automated trading, yield optimization, and AI-enhanced transaction strategies, as well as features like sniping, front-running, and monitoring.
Hey Anon was created by DeFi 2.0 legend Daniele Sesta, who previously led projects like Spell and Wonderland. It offers an AI interface based on natural language and simplifies operations like on-chain swaps, margin trading, and loans. Griffain, on the other hand, is a general DeFAI platform that allows users to perform on-chain tasks like token issuance, airdrops, and automated trading via natural language, drastically reducing DeFi complexity.
As an Ethereum L2 focused on DeFi, Mode Network invested heavily in AI agents even before they became widely adopted. It has incubated top-tier AI agent protocols such as Modius, ARMA, and AmpliFi, all of which drive the automation of DeFi strategies.
Meanwhile, ASYM specializes in DeFi automation and agent infrastructure, using on-chain data and social media to track market trends and achieve 3–4x actual trading returns.
Cod3x helps users quickly build automated trading agents through no-code tools and AlloraNetwork’s machine learning models. Its flagship agent, BigTonyXBT, has delivered impressive performance and seen a rapid rise in market cap. In contrast, Slate focuses on on-chain automated trading, integrating data from platforms like Telegram and Discord. With features like MEV protection and efficient routing, it is tailored for professional traders.
However, for now, DeFAI products still struggle with user acquisition, and the incentives for users to adopt them remain insufficient. No project has yet demonstrated a strong user base or DAU. These products still need to attract new users through wealth effects, such as helping them earn more.
Summary
Since Truth Terminal received a $50,000 donation in Bitcoin from Marc Andreessen and launched $GOAT in October 2024, the AI Agent sector has witnessed a shift from an asset-issuance frenzy to value consolidation. Initially, narrative, team pedigree, and product viability were paramount in driving interest and valuation. However, as the market matures, the focus is increasingly shifting towards protocol revenue and user adoption as key metrics for assessing value. DeepSeek’s technological breakthrough, demonstrating large language models’ ability to enhance reasoning through pure reinforcement learning, has injected fresh momentum into the sector. Future growth areas are anticipated in multi-agent systems, DeFAI, and upgrades to key frameworks and emerging infrastructure projects. In the end, sustainable profits, real user demand, and innovation will decide which projects survive, marking the AI Agent sector’s evolution from narrative-driven hype to a maturity phase grounded in value validation by data and revenue.
This report is published as a collaboration between HTX Research and HTX Ventures.
— — — — — — — — — — — — — — -
About Us
HTX Ventures, the global investment division of HTX, integrates investment, incubation, and research to identify the best and brightest teams worldwide. With a decade-long history as an industry pioneer, HTX Ventures excels at identifying cutting-edge technologies and emerging business models within the sector. To foster growth within the blockchain ecosystem, we provide comprehensive support to projects, including financing, resources, and strategic advice.
HTX Ventures currently backs over 300 projects spanning multiple blockchain sectors, with select high-quality initiatives already trading on the HTX exchange. Furthermore, as one of the most active FOF (Fund of Funds) funds, HTX Ventures invests in 30 top global funds and collaborates with leading blockchain funds such as Polychain, Dragonfly, Bankless, Gitcoin, Figment, Nomad, Animoca, and Hack VC to jointly build a blockchain ecosystem.
Company Website: https://www.htx.com/en-us/ventures
References:
https://x.com/shawmakesmagic/status/1884376511391674742
https://x.com/diego_defai/status/1881257334241890407?s=46
https://x.com/WheatiesSOL/status/1873727022427701355
https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf
