Walt plans to let traders execute perpetual futures orders through Mira, an AI layer that uses ChatGPT and Claude, according to details of the planned setup. Under the design, order execution would be off by default. AI agents would be confined to a dedicated trading account that users choose to fund.
The arrangement means a user could ask an AI model to place a perpetuals trade, but nothing would happen automatically until the user turns execution on. The dedicated account acts as a separate pot of money — one the user tops up only if they want the agent to trade.
What the Mira setup actually does
Mira sits between the user and Walt's perpetuals market. ChatGPT and Claude are the two models named in the plan. A trader would interact with Mira, and the AI would translate that into an order for the dedicated account.
The default-off execution setting is the central safety feature. A user can set up the account and talk to the AI without any order firing. Turning execution on is a deliberate step, not a side effect of signing up.
Perpetuals are leveraged contracts that track an underlying asset and never expire. They're a high-risk product even without an AI in the loop. Walt's plan keeps the AI on a short leash by design: separate account, opt-in execution, no access to the user's main balance.
Why the default matters
Default settings decide how most people use a product. An execution default of off means the AI can't trade by accident, can't trade because of a misread prompt, and can't trade because a model hallucinated an order. The user has to fund the dedicated account and switch execution on.
That's a different posture from tools that connect an AI directly to an existing brokerage balance. Walt's plan requires a second account and a manual toggle. The friction is the point.
The dedicated account also caps the damage. If an AI agent goes wrong, the loss is limited to what the user put into that account. The rest of the user's funds stay outside the agent's reach.
The two-model approach
Using both ChatGPT and Claude gives Mira two different underlying models. The plan doesn't say whether users pick a model or whether Mira routes between them. It also doesn't specify which tasks each model handles.
Two models can cross-check each other, or one can serve as a fallback when the other is unavailable. Either way, the user-facing result is a single AI layer that can call on either model.
What the plan doesn't include is any statement on latency, fees, or which perpetuals markets will be available at launch. Those details aren't in the facts.
What's still missing
Walt hasn't given a launch date. The plan describes the structure — AI perpetuals through Mira, execution off by default, a dedicated funded account — but not the rollout schedule.
There's also no word on how the dedicated account is funded, whether it's a separate wallet or a sub-account, or what happens if a user turns execution on and then wants it off mid-trade. Those are operational questions that will matter to anyone who tries the feature.
For now, the plan is a blueprint. The next concrete step is Walt's launch announcement, which hasn't been made.
