Top 10 AI+Crypto Trends to Watch in 2025

Original title: Crypto x AI: 10 Categories We're Watching in 2025
Original author: @archetypevc, Crypto Kol
Original translation: zhouzhou, BlockBeats

Editor's note:This article discusses multiple innovative areas of crypto and AI in 2025, including agent-to-agent interaction, decentralized agent organization, AI-driven entertainment, generative content marketing, data market, decentralized computing, etc. The article explores how to use blockchain and AI technology to create new opportunities in multiple industries, promote privacy protection, AI hardware development and the application of decentralized technology, and pay more attention to how intelligent agents can bring breakthroughs in transactions, artistic creation and other fields.

The following is the original content (the original content has been reorganized for easier reading and understanding):

Interaction between agents

The default transparency and composability of blockchain make it an ideal platform for interaction between agents.

In this scenario, agents developed by different entities for different purposes can interact with each other seamlessly. There have been many experiments on agents sending funds to each other, launching tokens together, etc.

We are looking forward to seeing how the interaction between agents can be expanded, including the creation of new application areas such as new social places driven by agent interaction, as well as improving efficiency and solving some of the current cumbersome problems by improving existing enterprise workflows (such as platform authentication, verification, micropayments, cross-platform workflow integration, etc.).

—Danny, Katie, Aadharsh, Dmitriy

Decentralized Agent Organizations

Large-scale multi-agent coordination is another equally exciting area of research.

How do multi-agent systems work together to accomplish tasks, solve problems, and manage systems and protocols? In his early 2024 article "The Promise and Challenges of Crypto and AI Applications", Vitalik mentioned that AI agents could be used for prediction markets and adjudication. He actually believes that in large-scale applications, multi-agent systems have significant "truth" discovery capabilities and can achieve universal autonomous governance systems. We are very interested in the capabilities of multi-agent systems and the continued discovery and experimentation of forms of "swarm intelligence".

As an extension of inter-agent coordination, coordination between agents and humans is also an interesting design space—particularly how communities interact around agents, or how humans can be organized through agents for collective action. We hope to see more experiments with agents that aim to coordinate humans at scale. This will need to be coupled with some verification mechanisms, especially if some of the human work is done off-chain, but it could also lead to some very strange and interesting emergent behaviors.

—Katie, Dmitriy, Ash

Agent-Based Multimedia Entertainment

The concept of digital characters has been around for decades.

Hatsune Miku (2007) sold out 20,000-seat arenas, and Lil Miquela (2016) has over 2 million followers on Instagram. Newer, lesser-known examples include AI virtual streamer Neuro-sama (2022), who has over 600,000 subscribers on Twitch, and the anonymously debuted Korean boy band PLAVE (2023), which has over 300 million views on YouTube in less than two years.

As AI infrastructure develops, and blockchain is integrated into payment, value transfer, and open data platforms, we are excited to see how these agents become more autonomous, potentially unlocking a new mainstream entertainment category by 2025.

—Katie, Dmitriy

Generative/Agent Content Marketing

In the first case, the agent itself is the product, while in the other scenario, the agent complements an existing product. In the attention economy, maintaining a constant flow of engaging content is critical to the success of any idea, product, company, etc.

Generated/agentic content is a powerful tool for teams to ensure a scalable, 24/7 content creation pipeline. The concept has been accelerated by discussions around what differentiates memecoins from agents. Agents provide a powerful means of distribution for memecoins, even if those memecoins are not fully “agentic” yet (but may become so).

As another example, games increasingly need to be more dynamic to keep users engaged. One of the classic ways to create game dynamics is to foster user-generated content; fully generated content (from in-game items to NPCs to fully generated levels) may be the next stage in this evolution. We are curious to see how agents will push the boundaries of traditional distribution strategies in 2025.

—Katie

Next Generation Art Tools/Platforms

In 2024, we launched IN CONVERSATION WITH, an interview series with crypto artists in music, visual art, design, curation, and more. A key observation I made from this year’s interviews is that many artists interested in crypto are often also interested in cutting-edge technologies and want to incorporate these technologies into their own artistic practices, in other words, AR/VR objects, code-based art, and live coding.

Generative art in particular has a natural synergy with blockchain, making it even more clear as a potential foundational platform for AI art. It is extremely difficult to properly display these art forms in traditional platforms. ArtBlocks provides a vision for how blockchain can be used to display, store, monetize, and preserve digital art in the future - improving the overall experience for artists and audiences. Beyond display, AI tools even expand the ability of ordinary people to create their own art. It will be very interesting to see how blockchain expands or supports these tools in 2025.

—Katie

Data Market

In the 20 years since Clive Humby coined the phrase “data is the new oil,” companies have taken strong steps to hoard and monetize user data. Users are gradually realizing that their data is the foundation on which these multi-billion dollar companies are built, but they have little control over how their data is used, nor do they share in the profits it brings.

The accelerated development of powerful AI models has made this contradiction even more acute. If addressing user exploitation is part of the data opportunity, then another important issue is addressing the shortage of data supply, as increasingly larger and more powerful models are consuming the easy oil fields of public Internet data and require new sources of data.

How to use decentralized infrastructure to transfer control of data from companies back to the source of data (users) is a huge design space involving innovative solutions in multiple fields. The most pressing issues include: where data is stored and how privacy is maintained during storage, transmission, and computation; how to objectively benchmark, filter, and assess data quality; what mechanisms do we use for attribution and monetization (especially when value needs to be traced back to the source after inference); and what coordination or data retrieval systems do we use across a diverse model ecosystem.

As for solving the supply bottleneck, it’s not just about replicating Scale AI through tokens, it’s more about understanding where we can gain advantages with the help of technology tailwinds, and how to build competitive solutions around scale, quality, or better incentive (and filtering) mechanisms to produce higher value data products. Especially when the demand side is mostly from web2 AI, how to combine smart contract enforcement mechanisms with traditional service level agreements (SLAs) and tools is an important area to focus on.

—Danny

Decentralized Computing

If data is one foundational building block for the development and deployment of AI, then computing power is another. In the past few years, traditional large data centers with unique access - including control of space, energy, and hardware - have dominated the trajectory of deep learning and AI, but this landscape is beginning to be challenged by physical limitations and the advancement of open source.

The v1 compute version in decentralized AI looks like a replica of web2 GPU clouds, with no real supply advantage (either hardware or data centers) and a lack of organic demand. In v2, we are starting to see some outstanding teams building a complete technology stack based on heterogeneous high-performance computing (HPC) supply, involving capabilities in coordination, routing, pricing, etc., combined with some proprietary features to attract demand and deal with marginal compression, especially in the inference stage. Teams are also starting to diverge in different use cases and go-to-market strategies (GTM), with some focusing on efficient inference routing integrating compilation frameworks on diverse hardware, while other teams are pioneering distributed model training frameworks on the compute networks they build.

We are even starting to see the emergence of an AI-Fi market with new economic primitives that turn compute and GPUs into yield assets, or use on-chain liquidity to provide data centers with another source of capital to acquire hardware. A major question here is to what extent decentralized AI (DeAI) will develop and deploy on the decentralized computing track, or, as in the storage space, whether the gap between ideology and practical needs will never be bridged, preventing the idea from realizing its full potential.

—Danny

Compute Accounting Standards

Associated with the incentives of decentralized high-performance computing networks, a significant challenge in coordinating heterogeneous computing resources is the lack of a unified standard for accounting for these compute resources. AI models uniquely add multiple complexities to the HPC output space, from model variants and quantization to adjusting the level of randomness through the model's temperature and sampling hyperparameters. In addition, AI hardware can introduce further complexity through the diversity of GPU architectures and different versions of CUDA. Ultimately, this leads to the need for how to account for models and compute market power when computing cross-platform in heterogeneous distributed systems.

Due in part to a lack of standards, we have seen multiple cases across web2 and web3 where models and compute markets have failed to accurately account for the quality and quantity of their compute power. This has resulted in users having to audit the true performance of these AI layers by running their own comparative model benchmarks and rate-limiting the execution of the compute market’s proof of work.

Given that a core tenet of crypto is verifiability, we expect the intersection of crypto and AI to be more easily verifiable than traditional AI in 2025. Specifically, it will be important for average users to be able to compare various aspects of a given model or cluster, especially those features that define the output, in order to audit and benchmark the performance of the system.

—Aadharsh

Probabilistic Privacy Primitives

In the article “The Promise and Challenges of Crypto and AI Applications”, Vitalik mentioned solving the unique challenges between crypto and AI:
“In cryptography, open source is the only way to make things truly secure, but in AI, the openness of models (and even their training data) greatly increases their vulnerability to adversarial machine learning attacks.”

While privacy is not a new research direction in the blockchain field, we believe that the popularity of AI will continue to accelerate the research and application of privacy cryptographic primitives. This year, privacy-enhancing technologies such as ZK, FHE, TEE, and MPC have made significant progress, with application scenarios including general applications such as private shared states for computation on encrypted data. At the same time, we have also seen centralized AI giants like Nvidia and Apple use proprietary TEEs for federated learning and private AI reasoning to keep hardware, firmware, and models consistent across systems.

With this in mind, we will be keeping a close eye on how to maintain privacy for random state transfers over heterogeneous systems, and how they can accelerate the development of real-world decentralized AI applications - from decentralized private reasoning to storage/access pipelines for encrypted data to fully sovereign execution environments.

—Aadharsh

Agent Intent and Next-Gen User Transaction Interfaces

The closest use case for AI agents is to use them to conduct autonomous transactions on-chain on our behalf. Admittedly, there has been a lot of fuzzy talk over the past 12-16 months about what "intentions", "agent behaviors", "agent intent", "solvers", "agent solvers", etc. are, and how they differ from the more traditional "bots" developed in recent years.

Over the next 12 months, we expect to see increasingly complex language systems combined with different data types and neural network architectures pushing the overall design space forward. Will agents continue to trade using the same on-chain systems we use today, or will they develop their own independent trading tools/methods? Will large language models (LLMs) continue to serve as the backend for these agent trading systems, or will they be replaced by other systems? At the interface level, will users begin to trade using natural language? Will the classic "wallet as browser" theory finally come to fruition?

—Danny, Katie, Aadharsh, Dmitriy

「Original link」