This is a segment from The Drop newsletter. To read full editions, subscribe.
US-based startup Axal is combining crypto trading with AI agents. They’re starting with automated trades through their Autopilot tool, which lets users set a risk tolerance level and other restrictions to automate trading with crypto and stablecoins.
The team told me in an interview that the tool currently uses Ethereum L2 chains Base, Optimism and Arbitrum on the backend, with plans to add Solana in the coming weeks.
Autopilot users can fund their accounts via crypto wallets, or by using an email and funding with Venmo. Once personal trading preferences and settings are set up, Autopilot’s AI agents can get to work.
There’s a wide range of functionality with Autopilot, but it’s more newbie-friendly and ideal for non-technical crypto fans who like crypto but aren’t day-traders or DeFi experts. Autopilot can collect stablecoin yields, engage with DeFi tools, and trade BTC, ETH, memecoins and other tokens. Users can also make a custom index of specific tokens for the AI agents to trade.
It’s not an intimidating tool, and it seems like there are enough parameters that can be set to ensure there will be no major surprises. But like with any AI tool, hallucinations are still possible.
Axal’s team is aware of this flaw, but they see the blockchain as one way to verify that requested tasks are actually completed per instructions.
“You can literally look onchain that this thing that you wanted this agent to do, it actually did,” Axal Head of Strategy and Operations Ari Santos said.
Under the hood, Axal’s also using zero-knowledge co-processors and Optimistic oracles to verify information.
“If a model hallucinates, it still has collateral at stake and it gets slashed. Whoever’s running the model, or the model itself, if it has very low accuracy, it’s just taking that risk,” Axal founder Ashlan Ahmed told Blockworks, explaining that hallucinating models are essentially punished for mistakes.