Written by: Haotian
Many people still do not understand why I have been advocating that AI framework standard projects should move towards 'chainization.' Perhaps during the previous two bull and bear cycles, Chain infrastructure carried too many expectations, and now that we have finally entered the 'application' era of AI Agents, people feel a sense of fear towards 'chains.' However, for AI Agents to have more reliable autonomous decision-making and intercommunication, they must inevitably trend towards 'chainization.'
Currently popular frameworks like ELIZA, ARC, and Swarms are still basically in the 'concept stage'; this stage cannot be falsified to zero, nor can it be confirmed to explode. They are essentially in a phase where their valuation cannot be quantified. This is the first hurdle for issuing assets on GitHub; finding the possibility of practical application for the outlined frameworks and visions is necessary to gain the market's consensus.
Upon closer examination of frameworks like ELIZA, ARC, and Swarms, whether optimizing the performance of a single AI Agent to the extreme or creating collaborative frameworks for multiple AI Agents, they essentially need to establish a traceable logic and rules for AGI large model API calls.
After all, the data is off-chain, the reasoning process is difficult to verify, the execution process is opaque, and the execution results are uncertain.
From a short-term perspective, TEE provides a low-cost, highly feasible off-chain Trustless solution that can accelerate the integration of AGI applications into the autonomous decision-making processes of AI Agents. From a longer-term perspective, a set of 'on-chain consensus' is also needed to assist in making it more reliable.
For example, ELIZA wants to build an autonomous private key custody solution for AI Agents based on its framework, utilizing the TEE security remote authentication capability of @PhalaNetwork to ensure that the execution code of the AI-Pool is not tampered with before calling the private key for signing. However, this is just the first small step of TEE's role in the direction of AI Agents.
If we can place complex preset execution logic into the Agent Contract and have Validators of the Phala chain collectively participate in validation, a chain based on consensus constraints of TEE execution details will be established. At that time, the demand for TEE driven by AI Agents will initiate a positive feedback loop empowering the chain.
The logic makes sense. TEE can ensure that private keys are not visible, but how the private keys are invoked, based on what preset rules, and how risk control emergency responses are triggered, etc. can be short-term addressed by open-source code repositories for transparency. However, in the long run, isn’t it all reliant on a decentralized validation consensus for real-time verification?
Therefore, 'chainization' can accelerate the transition of AI Agent frameworks into the practical application stage, and it can also bring new incremental opportunities alongside Crypto infrastructure.
The direction is already very clear. For most people, finding and being bullish on the earliest chain-based AI Agent frameworks and the oldest chains supporting AI Agents is the alpha under the new trend of AI Agents.
