The following three projects, currently in the testing phase, are all based on the underlying logic of 'providing core capabilities for AI agents through blockchain' rather than simply labeling. Their common characteristic is that without blockchain, the core functions of AI agents (autonomous collaboration, trusted payment, data rights confirmation) would be completely unachievable.
1. Fetch.ai (FET): A decentralized AI agent collaboration network
The core of Fetch.ai is not the 'AI model', but rather it has built a 'collaborative market without intermediaries' for AI agents. The role of its blockchain is to assign each AI agent a unique 'digital identity' and programmable wallet, allowing agents to autonomously complete the entire process of 'discovering needs - negotiating cooperation - executing tasks - settling rewards'.
• Implementation testing scenario: Intelligent scheduling in the logistics industry. Multiple AI agents (representing freight companies, warehouses, and fleets) can autonomously match transport demands on the chain, negotiate freight and time through smart contracts, and automatically settle using FET tokens upon task completion, without any manual intervention. The blockchain here resolves 'trust issues'—all collaboration rules and transaction records are immutable, avoiding default risks among agents.
• Key difference: The collaboration of AI agents is 'decentralized', not relying on any centralized server for matching; the demand for tokens entirely comes from service settlement among agents, rather than external staking incentives.
2. SingularityNET (AGIX): Decentralized trading and invocation network for AI models.
The core of SingularityNET is 'allowing AI models to be freely invoked like DeFi protocols', and the blockchain serves as the 'trust infrastructure' to achieve this goal. Developers can deploy their AI models (such as image recognition and natural language processing) on the chain and set invocation fees; other AI agents or users can directly pay fees using AGIX tokens to invoke these models, with the entire process executed automatically by smart contracts.
• Implementation testing scenario: Cross-model AI agent collaboration. A 'medical diagnosis AI agent' can autonomously invoke the 'medical image recognition model' and 'medical record analysis model' on SingularityNET, integrating the results of both models to generate diagnostic recommendations, with invocation fees settled in real-time using AGIX. The blockchain here resolves 'transparency and rights confirmation'—the invocation records of the models and fee distribution (developer shares) are fully traceable on the chain, preventing model theft or fee interception.
• Key difference: The blockchain carries 'trading and rights confirmation of AI models', rather than simply token transfers; the core capability of AI agents (multi-model integration) relies on the model market on the chain, with the blockchain being the sole channel for model invocation and payment.
3. Ocean Protocol (OCEAN): Trusted data interaction network for AI agents.
The entry point of Ocean Protocol is 'data'—providing 'trusted data sources and trading markets' for AI agents, with the core role of its blockchain being 'data rights confirmation' and 'incentive mechanisms'. Users can upload their data (after anonymization) to the chain, setting rules and fees for data usage; when AI agents need training data, they can pay with OCEAN tokens through smart contracts to obtain data usage rights, while data contributors can earn token dividends.
• Implementation testing scenario: AI-driven prediction market data feeding. A 'crypto market prediction AI agent' can autonomously filter and purchase trusted historical trading data and public opinion data on Ocean Protocol to train predictive models. The prediction results generated by the model are then fed back to the market through smart contracts, with users required to pay OCEAN to view the results, while the agent continues to purchase new data to iterate the model with the earnings. The blockchain here resolves 'data trustworthiness and incentives'—the sources, usage records, and revenue distribution of the data are fully traceable on the chain, ensuring that AI agents obtain 'clean data', while incentivizing users to continuously contribute high-quality data.
• Key difference: The blockchain is not an auxiliary tool, but a prerequisite for 'data becoming tradable assets'; the model iteration of AI agents entirely relies on the trusted data market on the chain, with ownership and revenue rights of the data clearly defined through the blockchain.
The testing progress of these three projects directly reflects the implementation path of 'AI agents' from 'theory' to 'practicality'—their underlying blockchain respectively addresses the three core pain points of 'collaborative trust', 'capability invocation', and 'data trustworthiness' of AI agents. It also serves as the best reference standard currently distinguishing 'true integration' from 'pseudo-concept' projects.
