Author: Iraklis A, CoinTelegraph; Compiled by: Baishui, Golden Finance

The rise of AI-driven crypto-agents follows a familiar trajectory that reflects the initial boom, bust and recovery of projects in the ICO era. Just as early blockchain companies flourished on hype before they matured into sustainable ecosystems, the current wave of AI agent projects is experiencing a rapid market shift.

A new report from HTX Ventures and HTX Research says investors are becoming more cautious as competition in the industry increases, liquidity is diversified and use cases that many projects are difficult to define. However, as the industry moves out of the speculative stage, AI-driven crypto proxy is expected to develop a sustainable business model based on real utility.

The initial wave of crypto Agent projects in 2024 was driven by a craze for AI projects. Amid the influence of Marc Andreessen’s donation of $50,000 in Bitcoin in October 2024 and the successful token offering earlier this year, many AI agent projects entered the field in the first quarter of 2024 and quickly diluted liquidity in the first quarter of 2025. As with any emerging industry, early hype doesn’t always translate into long-term viability, and the crypto AI agent industry then enters a cooling-off period.

The market sector is now entering a more mature stage, with the focus shifting from speculative excitement to revenue generation and product performance. In this ever-changing environment, winners will be those who can generate stable revenue, pay the cost of running AI models, and provide tangible value to users and investors.

AI agent application emphasizes the implementation and commercialization of the technology in the real world, especially in areas such as automatic trading, asset management, market analysis and cross-chain interaction. This approach is consistent with multi-agent systems and DeFAI (Decentralized Finance + AI) initiatives such as Hey Anon, GRIFFAIN, and ChainGPT.

Recent research highlights the advantages of multi-agent systems (MAS) in portfolio management, especially in cryptocurrency investment. Projects such as Griffain, NEUR, and BUZZ have demonstrated how AI can help users interact with DeFi protocols and make informed decisions. Unlike the single agent AI model, multi-agent systems utilize collaboration among special agents to enhance market analysis and execution. These agents operate in teams, such as data analysts, risk assessors and transaction execution departments, each agent trained to handle specific tasks.

MAS framework also introduces inter-agent communication mechanisms, and agents in the same team improve predictions through collective learning, fromAnd reduce errors in market trend analysis. The next phase of DeFAI may involve deeper integration of decentralized governance models, where multi-agent systems participate in protocol management, financial optimization, and on-chain compliance execution.

DeepSeek-R1 is a breakthrough in AI proxy technology, and this innovation challenges traditional AI training methods. Unlike previous models, DeepSeek-R1 relies on supervised fine-tuning (SFT), followed by reinforcement learning (RL), while DeepSeek-R1 adopts a different approach, optimized entirely through reinforcement learning without the initial supervised phase. This shift significantly improves reasoning capabilities and adaptability, paving the way for more complex AI-driven encryption agents.

To understand this paradigm shift, consider two different learning methods. In traditional SFT and RL models, students first learn from the exercise book, practice questions with fixed answers (SFT), and then receive tutoring to refine their understanding (RL). In contrast, in the DeepSeek-R1 model (pure reinforcement learning), students take the exam directly and learn through trial and error. This approach allows students to improve dynamically based on feedback rather than relying on predefined answers.

Utilizing DeepSeek-R1’s pure RL model, AI agents can learn through trial and error under real-world conditions and adjust their strategies dynamically based on instant feedback.

This approach is more adaptable and is particularly useful for multi-agent AI systems in DeFi, because real-time market fluctuations require agents to make autonomous, data-driven decisions. For example, an AI-powered agent can monitor liquidity pools, detect arbitrage opportunities, and optimize asset allocation based on real-time market conditions. These agents can quickly adapt to market volatility and ensure more efficient capital allocation.

iDEGEN, launched in late November 2024, is the first crypto AI proxy built on DeepSeek R1. This integration of DeepSeek's R1 model emphasizes how cryptographic AI agents inherit this enhanced inference capability to compete with other mature AI models at extremely low cost.

The shift to RL-driven multiagent AI in DeFi automation highlights why closed-source AI models, such as OpenAI’s GPT-based systems, are becoming an unsustainable expense. Since workflows typically require more than 10,000 tokens to be processed per transaction, closed AI models can incur huge computational costs, limiting scalability. In contrast, open source RL models like DeepSeek-R1 allow for customization for DeFi applicationsCost-effective and efficient AI development.

The key to the long-lasting prosperity of this field lies in continuous innovation, adaptability and cost-effectiveness. Open source AI models like DeepSeek-R1 are lowering the barrier to entry, allowing blockchain-native startups to develop dedicated AI solutions. At the same time, advances in DeFAI and multi-agent systems will drive the long-term integration between AI and decentralized finance.

The conclusion is clear: the project must prove its value beyond hype. Those who develop sustainable economic models and leverage cutting-edge AI advances will define the future of the smart blockchain ecosystem. The ICO era for crypto agents is developing, and the next wave of winners will be those who can turn innovation into long-term viability.