Crypto infra has spent a decade moving toward user control: → Self-custody wallets → Decentralized exchanges → Non-custodial lending Now that AI agents are emerging at an insane pace, I'm seeing that same logic extend to them. I realized that CT focuses heavily on agent execution (spending, trading), but barely mentions who controls the memory. If an agent's memory isn't verifiable, how can you know it hasn't been tampered with or silently updated? Some teams like @WalrusProtocol are working on verifiable storage solutions to address this, but most agent infra still treats memory as an afterthought. If we wait until agents are managing real capital at scale to fix this, we're repeating the CeFi mistakes all over again. This topic is highly overlooked, so I'll do a deep dive on it soon.