web3 agents: the new meta

The future of AI is agents, say OpenAI, Google and Microsoft, as well as founders of Coinbase, NEAR and EigenLayer. For both web2 and web3 world this will be the most important technological wave of this decade. Let's explore what opportunities onchain agents create and why did they get to almost $4B market cap in just one month? what's next for web3 Agents beyond the hype and speculation?

Below I'm exploring web3 AI agents: what they are, how did they get to $3B mkt cap in <1 month, how they function and what opportunities exist.

TL;DR

— AI agents will be the dominant narrative of 2025: Following the trajectory of previous crypto waves, AI agents are positioned to become the next major trend. — Growing ecosystem with ginormous potential: Despite numerous existing projects, there's significant room for growth and innovation in the space. — Real utility sets agents apart: Unlike NFTs and memecoins, AI agents provide practical value through their ability to perform useful work and generate actual cashflows. — Value-backed tokens: Agent tokens will be supported by real economic value, making them potentially less risky and more attractive for retail investors compared to purely speculative assets. — Early-stage opportunity: Infrastructure is still nascent, distribution channels need development and the design space remains largely unexplored.

web3 market cycles

AI agents are slowly but surely automating and improving efficiency across most knowledge tasks: from entertainment and social media to business productivity, marketing, finance, investment, healthcare and education.

Blockchain and web3 offer AI agents unique capabilities allowing for trustless composability, verifiability and programmable contracts for agentic economy.

Unlike memecoins, NFTs and ICOs, AI agent tokens can offer utility, marking a shift towards more substantive value propositions. Market narratives come and go, real products stay. Specifically, web3 agents are different from any existing AI application in two main properties, they have:

1. Autonomy: An agent operates in a decentralized computing environment.

2. Economy: For users and investors, participating in the agent’s economy is much more transparent and straightforward since it’s programmable, blockchain-based, and open.

Right now, we’re seeing early experiments reminiscent of Ethereum circa 2016. But by mid-to-late next year, we’ll likely see agents handling complex tasks, automating processes, and generating decent cash flow. Replacing workers with bots is inevitable — but hey, at least you’ll get to own their tokens.

web3 agents: current state

Currently, most Web3 AI agents operate as prompted large language models (LLMs) that generate text-based content on platforms like Twitter and Discord. These agents are designed to emulate characters or personas, often engaging users through social media interactions.

The primary functionalities include:

• Trading: Agents typically have a coin or token assoiated with it, that users can freely trade on various onchain and centralized exchanges.

• Content generation: Producing text responses, tweets or messages based on predefined prompts or fine-tuned models.

• Role-playing: Acting as specific characters or entities to engage users in interactive dialogues.

• Community engagement: Allowing users to influence agent behavior through prompts or feedback, creating a sense of participation.

The economic model of current Web3 AI agents involves tokens associated with each agent. These tokens are:

• Speculative assets: Primarily traded on exchanges, their value driven by market speculation rather than intrinsic utility.

• Lack of trustless integration: Tokens are not inherently linked to the agent's operations in a decentralized or verifiable manner.

• Absence of cash flows: Most agents do not generate revenue streams that benefit token holders, limiting long-term value.

Despite the decentralized ethos of blockchain, most current AI agents suffer from centralization issues:

• Centralized inference: Reliance on centralized services like Runpod or Lambda for model inference.

• Server dependency: Agent behavior, reasoning and memory are managed on central servers, creating single points of failure. It's typically a python script using Langchain, CrewAI or similar library that is centrally run on a single machine.

• Limited transparency: Central control over agent operations reduces transparency and trust among users and token holders. This opens a possibility for "rug pull" or agent creator making decisions on their own.

Design Space of web3 AI Agents

The design space for Web3 AI agents is vast, offering a range of possibilities.

Fun ↔ Utility Spectrum

Developers can position their AI agents along a spectrum from pure entertainment to practical utility, each appealing to different market niches.

Entertainment-focused agents prioritize user engagement through fun and interactive experiences. They may:

• Role-play characters: Act as dynamic personas on social media or virtual environments, engaging users through storytelling or humor.

• Host media content: Run podcasts, livestreams or virtual events, attracting audiences with unique AI-generated content.

• Community interaction: Encourage user participation by allowing the community to influence the agent's behavior or narrative.

Utility-focused agents are aimed at solving real-world problems, these agents offer tangible value by automating tasks and enhancing productivity. They may:

• Automate knowledge work: Perform tasks like coding, data analysis, market research or content creation.

• Provide services: Act as AI therapists, educators, customer support or social media managers.

• Enhance business operations: Optimize workflows, improve decision-making or offer insights through advanced analytics.

Agency ↔ Time-to-Market

There's a trade-off between developing highly autonomous agents with advanced features and bringing products to market quickly.

Fast & simple deployment:

• Quick launch: Simple agents can be developed and deployed rapidly using existing models and centralized infrastructure.

• Limited functionality: These agents may offer basic features and rely on single modalities like text interaction.

• Speculative appeal: By tapping into current trends, they can attract immediate attention and investment, often driven by speculative trading.

Complex, time-intensive development:

• On-chain governance: Implementing decentralized control mechanisms allows for community governance over agent parameters and development.

• Revenue sharing with token holders: Establishing economic models where agents generate income that benefits token holders.

• Community-owned agents: Creating agents that are collectively owned and managed by the community, enhancing engagement and sustainability.

• Infrastructure requirements: Developing these features requires building robust frameworks, decentralized inference systems and privacy-preserving computations, which take time and resources.

Memecoins ↔ Real Cash Flows

The economic models of AI agent tokens can range from speculative assets to tokens backed by real economic value.

Speculative memecoins gain value primarily through market hype and speculative trading, typically without underlying utility. While this approach can be risky, it serves as a bootstrap mechanism, helping to attract initial investment and community interest to new projects. However, it's important to note that relying solely on speculation can lead to market volatility and potential losses, which may undermine the long-term sustainability of the project.

However, reliance on speculation leads to market volatility and potential losses, undermining long-term sustainability.

Tokens with real cash flows provide tangible value to holders and contribute to a more sustainable ecosystem. By representing access to agent services, governance rights or revenue sharing, these tokens create a direct link between the agent's performance and token value. This alignment fosters a stable and growing ecosystem, as the token's worth is tied to the agent's utility and success. Furthermore, offering real economic benefits builds investor confidence, encouraging long-term participation and support for the project. This approach not only attracts serious investors but also helps create a more resilient and value-driven token economy in the AI agent space.

Future of Web3 AI Agents

I believe AI agents will likely become the dominant crypto narrative in 2025. However, there are still many unsolved problems and room for growth. Below, I'll try to structure the potential areas for improvement to make AI agents more useful and better overall.

Better Tokenomics

Developing robust and sustainable economic models is crucial for the long-term success of AI agents.

Revenue sharing is a key aspect of tokenomics for AI agents. Agents can generate income through services like content creation, coding or analysis, distributing profits to token holders. This aligns financial rewards with agent performance, encouraging community members to contribute to the agent's success.

Utility-backed value creates tangible benefits for token holders. Access tokens grant users entry to the agent's services, while governance rights allow token holders to influence development decisions, agent parameters and strategic directions.

Community ownership empowers users through decentralized control. Implementing governance mechanisms allows the community to collectively manage the agent. Multi-signature wallets and smart contracts provide secure and transparent methods for handling payments, revenue distribution and agent actions.

Mechanism design help build beneficial behavior among agents, creating an agentic economy where agents can transact, negotiate and form agreements autonomously. Smart contracts play a vital role in automating transactions and enforcing agreements without intermediaries.

Agent orchestration focuses on coordinating multiple agents to work together, enhancing overall capabilities and efficiency of the system.

True Decentralization

Overcoming centralization challenges is essential for building trust and ensuring the resilience of AI agents.

Inference and execution are crucial aspects of decentralizing AI agents. Decentralized inference distributes model processing across networks or edge devices, eliminating central server reliance. Privacy-preserving computation, using methods like zero-knowledge proofs, secures data and calculations. These approaches enhance system resilience against downtime and censorship, while ensuring verifiable execution for community auditing.

On-chain execution moves agent reasoning and decision-making onto the blockchain, significantly improving transparency and trustlessness. This involves implementing smart contract logic for agent operations. Additionally, leveraging edge computing and distributed systems allows for efficient and secure computation across decentralized networks.

Community governance plays a vital role in truly decentralized AI agents. By empowering token holders with collective decision-making capabilities, the risk of centralized control is reduced. Governance mechanisms ensure that the agent's actions align with the community's values and objectives, creating a more democratic and user-driven development process.

More Modalities and Better Tooling

Expanding the capabilities of AI agents and improving development tools will drive utility and adoption by the wider market.

Multimodal interaction is crucial, incorporating voice, video and other forms of communication to enrich user experience. Advanced communication enables agents to exchange machine-readable data, such as embeddings or model parameters, enhancing efficiency in agent-to-agent interactions.

Access to specialized models, such as time series analysis, can open up many more use cases for web3 agents. For example, they can become traders and DeFi strategy optimizers that manage and allocate capital on your behalf.

Tooling advancements are essential for the growth of AI agents. Robust frameworks for spawning and managing agents simplify the creation of complex functionalities. Integrating advanced logic, including reasoning, planning, self-critique and tool integration, allows agents to perform sophisticated tasks. Establishing interoperability standards and protocols facilitates seamless multi-agent collaboration, paving the way for more complex and efficient AI ecosystems.

Workflow orchestration systems play a crucial role in optimizing AI agent performance. These systems enable dynamic adaptation, adjusting workflows in real-time based on performance and changing conditions. They also facilitate optimal path selection, choosing the most efficient routes for task completion while balancing cost and quality. Encouraging interactions among agents can lead to emergent behaviors, fostering innovative solutions and capabilities. Additionally, designing fault-tolerant distributed systems ensures continued effective operation even if some agents fail.

Security remains a paramount concern in AI agent development. Addressing AI-specific security challenges, such as model stealing, prompt injection and data poisoning attacks, is crucial for building trust and reliability. Equally important is ensuring auditability and interpretability, making agents' decision-making processes transparent and auditable for compliance and trust. These measures are essential for the widespread adoption and integration of AI agents in various domains.

Agent Experience

Like many aspects of crypto, using web3 agents isn't trivial and often requires sophisticated knowledge. However, this is set to change.

web3 agents will significantly improve the user experience by adding dynamic generative interfaces that adapt to individual user preferences. An agent will generate UI elements in real-time, using familiar and convenient UI patterns personalized for each user.

Voice commands, 3D avatars, AR features that enhance engagement and make it more natural. Seamless connection or even integration inside crypto wallets and other tools enabling frictionless interactions. Persistent memory systems allow for more personalized and relevant interactions.

24/7 availability with autonomous action capabilities when authorized, such as monitoring token prices and executing trades at specific thresholds, scheduling regular social media posts or automatically responding to Discord messages based on predefined triggers. The agent can also proactively notify users about important events like governance proposals, upcoming token unlocks or suspicious wallet activities that require attention.

This transformation will elevate agents from basic chatbots to truly useful agents that can entertain or help users solve problems at scale.

Agents will replace human workforce

I believe AI agents represent a fundamental shift in the global workforce. Over time, more jobs will be automated, with AI agents performing tasks faster, cheaper, and more effectively—ultimately leading to greater prosperity. AI agents combined with tokenization offer users a chance to own a share of this post-labor economy, ensuring the benefits of automation are distributed rather than concentrated in the hands of a few tech giants.

While today's web3 agents closely resemble memecoins with their speculative properties, I expect them to gradually develop genuine utility and practical value. You can also read my previous article where I talk more practically about where the demand for agent services will come from.

Cybernetic economy is near. The swarm awakens.

Please, reach out to me if you're building onchain AI agents or agent infra and looking for advice, investments, connections.