Original: Archetype
Compiled by: Yuliya, PANews
In the rapidly evolving landscape of artificial intelligence and Block technology, the convergence of these two domains is giving rise to exciting innovative possibilities. This article delves into the ten most important areas worth watching in 2025, from intelligent agent interactions to decentralized computing, from the transformation of data markets to breakthroughs in privacy technology.
1. Agent-to-Agent Interaction
The inherent transparency and composability of Block technology make it an ideal foundational layer for agent-to-agent interaction. Smart agents developed by different entities, serving different purposes, can seamlessly interact on the Block. Some noteworthy experimental applications have already emerged, such as agents transferring funds and jointly issuing tokens.
The future development potential of agent-to-agent interaction lies in two main aspects: first, creating entirely new application domains, such as novel social scenarios driven by agent interactions; and second, optimizing existing enterprise-level workflows, including platform authentication and verification, micro-payments, and cross-platform workflow integration - traditionally more cumbersome processes.
aethernet and clanker jointly issue tokens on the Warpcast platform
2. Decentralized Intelligent Agent Organizations
Large-scale multi-agent coordination is another exciting research area. This involves how multi-agent systems can collaborate to accomplish tasks, solve problems, and govern systems and protocols. In his early 2024 paper "The Prospects and Challenges of Cryptocurrency and AI Applications", Vitalik mentioned the potential of using AI agents in prediction markets and arbitration. He believes that from a macro perspective, multi-agent systems show significant potential in "truth" discovery and autonomous governance systems.
The industry is continuously exploring and experimenting with the capability boundaries of multi-agent systems and various forms of "collective intelligence". As an extension of agent-to-agent coordination, the coordination between agents and humans also constitutes an interesting design space, particularly in how communities interact around agents, and how agents can organize humans for collective action.
Researchers are particularly focused on agent experiments where the objective function involves large-scale human coordination. Such applications require corresponding verification mechanisms, especially when humans are working off-chain. This human-agent collaboration may give rise to some unique and interesting emergent behaviors.
3. Intelligent Agent Multimedia Entertainment
The concept of digital personas has existed for decades.
- As early as 2007, Hatsune Miku was able to hold sold-out concerts in venues of 20,000 people;
- The virtual influencer Lil Miquela, born in 2016, has amassed over 2 million followers on Instagram;
- The AI virtual streamer Neuro-sama, launched in 2022, has accumulated over 600,000 subscribers on Twitch;
- The virtual K-pop group PLAVE, established in 2023, has garnered over 300 million views on YouTube in less than two years.
With the advancement of AI infrastructure and the integration of Block technology in payments, value transfer, and open data platforms, these intelligent agents are poised to gain greater autonomy by 2025 and may potentially create a new mainstream entertainment category.
Clockwise from top left: Hatsune Miku, Luna from Virtuals, Lil Miquela, and PLAVE
4. Generative/Intelligent Agent Content Marketing
In contrast to intelligent agents as products themselves, they can also serve as complementary tools for products. In today's attention economy, continuously producing engaging content is crucial for the success of any creative, product, or company. Generative/intelligent agent content is becoming a powerful tool for teams to ensure 24/7 uninterrupted content production.
The development in this area has been accelerated by the discussion around the blurring boundaries between Meme coins and intelligent agents. Even without fully achieving "intelligence", intelligent agents have already become a powerful means for Meme coins to gain traction.
The gaming domain provides another typical case. Modern games increasingly need to maintain dynamism to sustain user engagement. Traditionally, cultivating user-generated content (UGC) has been a classic method for creating game dynamism. Pure generative content (including in-game items, NPC characters, fully generated levels, etc.) may represent the next stage of this evolution. Looking ahead to 2025, the capabilities of intelligent agents will greatly expand the boundaries of traditional distribution strategies.
5. Next-Generation Art Tools and Platforms
The "IN CONVERSATION WITH" interview series launched in 2024 spoke with artists active in the cryptocurrency space and its periphery, including musicians, visual artists, designers, and curators. These interviews revealed an important observation: artists interested in cryptocurrencies often also focus on broader frontier technologies and tend to deeply integrate these technologies into the aesthetics or core of their artistic practices, such as AR/VR objects, code-based art, and live-coding art.
Generative art and Block technology have long had synergistic effects, making the potential of the latter as an AI art infrastructure more apparent. On traditional display platforms, it is extremely difficult to properly showcase these new art media. The ArtBlocks platform has demonstrated a future vision of using Block technology for digital art display, storage, monetization, and preservation, significantly improving the overall experience for both artists and audiences.
In addition to display functions, AI tools have also expanded the ability of the general public to create art. This democratization trend is reshaping the landscape of artistic creation. Looking ahead to 2025, how Block technology will expand or empower these tools will be an highly attractive area of development.
Excerpt from "In Conversation: Maya Man"
6. Data Markets
It has been 20 years since Clive Humby proposed the idea that "data is the new oil", and major companies have been taking strong measures to hoard and monetize user data. Users have become aware that their data is the foundation of these multi-billion dollar companies, but they have almost no control over how their data is used and cannot share in the profits it creates. With the rapid development of powerful AI models, this contradiction has become more pronounced.
The opportunities facing data markets have two aspects: one is to solve the problem of user data exploitation, and the other is to solve the problem of data supply shortage, as increasingly larger and better models are consuming the easily accessible "oil field" of public internet data and require new data sources.
Returning Data Power to Users
Leveraging decentralized infrastructure to return data power to users is a broad design space that requires innovative solutions across multiple domains. Some of the most pressing issues include:
- Where data is stored and how to protect privacy during storage, transmission, and computation;
- How to objectively evaluate, filter, and measure data quality;
- What mechanisms to use for attribution and monetization (especially in tracing value back to the source in inference);
- And what kind of orchestration or data retrieval systems to use in a diversified model ecosystem.
Supply Constraints
In addressing supply constraints, the key is not simply to replicate Scale AI's token-based model, but to understand where we can build advantages in a technology-favorable environment, and how to construct solutions with competitive advantages in scale, quality, or better incentive (and filtering) mechanisms to create higher-value data products. Particularly as the majority of demand still comes from Web2 AI, exploring how to integrate smart contract execution mechanisms with traditional service-level agreements (SLAs) and tools is an important research area.
7. Decentralized Computing
If data is a fundamental element of AI development and deployment, then computing power is another key component. The traditional large-scale data center model has largely dominated the development trajectory of deep learning and AI in recent years, leveraging its unique advantages in terms of site, energy, and hardware. However, physical constraints and the development of open-source technologies are challenging this landscape.
- The first stage (v1) of decentralized AI computing is essentially a replica of Web2 GPU cloud services, without real advantages on the supply side (hardware or data centers) and with limited organic demand.
- In the second stage (v2), some outstanding teams have built a complete technology stack on the basis of heterogeneous high-performance computing (HPC) supply, demonstrating unique capabilities in scheduling, routing, and pricing, while developing proprietary features to attract demand and address profit compression, especially on the inference side. Teams have also begun to differentiate in terms of use cases and market strategies, with some focusing on integrating compiler frameworks to achieve efficient inference routing across hardware, and others creating distributed model training frameworks on the computing networks they have built.
The industry has even begun to see the emergence of an AI-Fi market, with innovative economic primitives that convert computing power and GPUs into income-generating assets, or leverage on-chain liquidity to provide an alternative source of funding for data centers to acquire hardware.
The key question here is to what extent decentralized AI will develop and deploy on decentralized computing infrastructure, or whether, as in the storage domain, the gap between the ideal and actual demand will persist, preventing this concept from fully realizing its potential.
8. Computational Accounting Standards
One of the main challenges in the incentive mechanism of decentralized high-performance computing networks is the lack of a unified computational accounting standard. AI models have added multiple unique complexities to the output space of high-performance computing, including model variants, quantization schemes, and adjustable levels of randomness through model temperature and sampling hyperparameters. Additionally, AI hardware can produce different outputs due to differences in GPU architectures and CUDA versions. These factors ultimately require the establishment of standards to regulate how models and computing markets measure their computational capabilities in heterogeneous distributed systems.
Partly due to the lack of these standards, there have been several cases in the Web2 and Web3 domains in 2024 where model and computing markets have failed to accurately account for their computational quality and quantity. This has led users to have to run their own comparative model benchmarks and execute proof-of-work by throttling the rate of the computing market to audit the true performance of these AI layers.
Looking ahead to 2025, the intersection of cryptography and AI is expected to achieve breakthroughs in verifiability, making it more easily verifiable compared to traditional AI. For the average user, the ability to fairly compare various aspects of the defined model or compute cluster outputs is crucial, as it will aid in auditing and evaluating system performance.
9. Probabilistic Privacy Primitives
In "The Prospects and Challenges of Crypto and AI Applications", Vitalik pointed out a unique challenge in connecting cryptocurrencies and AI: "In cryptography, open-source is the only way to achieve true security, but in AI, the openness of models (and even their training data) greatly increases the risk of adversarial machine learning attacks."
While privacy is not a new field of blockchain research, the rapid development of AI is accelerating the research and application of privacy-preserving cryptographic primitives. Significant progress has been made in privacy-enhancing technologies in 2024, including zero-knowledge proofs (ZK), fully homomorphic encryption (FHE), trusted execution environments (TEEs), and multi-party computation (MPC), which are being used in general-purpose applications such as encrypted data computation and private shared state. Meanwhile, centralized AI giants like Nvidia and Apple are also using proprietary TEEs for federated learning and private AI inference, ensuring privacy while maintaining hardware, firmware, and model consistency across systems.
Based on these developments, the industry is closely watching the progress of privacy-preserving technologies in probabilistic state transitions, and how these technologies can accelerate the practical deployment of decentralized AI applications on heterogeneous systems. This includes aspects ranging from decentralized private inference to encrypted data storage/access pipelines, and fully sovereign execution environments.
10. Proxy Intentions and Next-Generation User Trading Interfaces
Over the past 12-16 months, there has been a lack of clear definition around concepts such as intent, proxy behavior, proxy intent, solutions, and proxy solutions, and how these differ from the traditional "robot" development in recent years. AI proxies autonomously conducting on-chain transactions is one of the closest use cases to being realized.
In the next 12 months, the industry expects to see more complex language systems combined with different data types and neural network architectures, advancing the overall design space. This raises several key questions:
- Will proxies use existing on-chain trading systems, or develop their own tools and methods?
- Will large language models continue to serve as the backend for these proxy trading systems, or will entirely new systems emerge?
- At the interface level, will users start using natural language for trading?
- Will the classic "wallet as browser" concept ultimately be realized?
The answers to these questions will profoundly shape the future development of cryptocurrency trading. As AI technology advances, proxy systems may become more intelligent and autonomous, better able to understand and execute user intent.
