Trading bots are not new. Nor is the idea of giving one its own crypto wallet. Connecting an LLM to the X API is also not novel. But a trading bot with its own wallet spewing a stream of 4chan-like slang, plus the ability to gamble on it via memecoins, is something new. With the help of a few notable backers, these components were enough for the Truth Terminal token (GOAT) to achieve escape velocity in the memecoin market.

While most 'AI' memes like GOAT are little more than rugs waiting to happen, Tee He He, launched by Flashbots and Nous Research, is a more interesting experiment. It's the first implementation of a provably autonomous AI Agent within a trusted execution environment (TEE) and an early example of how to build trustworthy and verifiable agents. While this particular meta will eventually fade, there's a clear product-market fit for agents to fulfill all kinds of intents. Indeed, crypto is accelerating toward a future with provably autonomous agents with the agency who do more than “shitpost.” 

This future means more functionality, longer context windows, increased chain of thought, and human level or above performance. The rise of agents could also make it significantly easier and faster to create products. Without the need to build a human-friendly UI for everything, developers can focus on the smaller components and how they connect to each other to form gestalt mechanisms and other emergent phenomena. Some models are already incredibly smart on their own, but the real question is if they will be able to self-configure into novel patterns that, ultimately, overcome a billion years of natural selection. 

In a more practical sense, as these increasingly autonomous agents get plugged into our systems, it raises questions about payments, reliability, and data confidentiality. How will crypto be able to trust agents? How can participants be sure an agent will perform as expected? How will users be able to protect sensitive data? At the end of the day, an agent is just software running on hardware, which means if it's running on traditional servers, these agents will typically store info like passwords and private keys, leaving anyone interacting with them effectively at the creator's mercy. 

The solution, as demonstrated by Tee He He, is trusted execution environments. TEEs create secure enclaves for processing sensitive data, allowing applications to execute securely in an isolated environment. TEEs work by only allowing certain, verifiable code to execute within it, and via a remote attestation mechanism, it can prove the integrity of what's running inside it, and the content & runtime state are hidden from node operators. The upshot is anyone can confirm that an agent (if in a TEE) is operating strictly according to the code without tampering or interference. 

This is precisely why Oasis is developing Runtime Offchain Logic (ROFL), a generalized framework for verifying offchain compute. ROFL enables arbitrary applications, like agents, to function verifiably through TEEs with authenticated & encrypted communication with smart contracts. It's also possible to solve the issue of private key custody for agents via another TEE-enabled piece of infra, Oasis' confidential EVM Sapphire. As agents need a private key to operate a wallet, using Sapphire makes it possible to store agent keys onchain and then build logic within a smart contract to retrieve keys, sign messages cross-chain, etc.

Together, ROFL and Sapphire make it possible to create these verifiably independent agents, where there's some certainly of not being rugged. They also enable a layer of confidentiality. So, in the case of LLM-powered applications, frameworks like ROFL safeguard sensitive inputs/outputs (e.g. agentic doctor) and, in some cases, protect the models themselves. The human-optional web is coming, and TEEs provide the verifiable/privacy features that agents need. Full Intel TDX support is on testnet and will soon be in production on ROFL. Learn more here. 

‍