At @finalitycap, we think “Plaid for AI memory” is a massive opportunity. Imagine every app or AI model instantly knowing your context, preferences, interests, and history This is huge for UX + costs 🧵👇 Props to @binji_x @nikunj @ashleymayer @levie for sparking great ideas:

2/ Today, every AI model keeps your chats siloed. ChatGPT doesn’t know your Claude or Gemini history or even your identity 🪪 Users waste time and money re-explaining themselves. It’s not just frustrating, it’s also costly 💸

3/ Users waste tokens and money by re-entering context on LLMs like personal info, prefs, etc. in every session Since LLMs charge by tokens, this inefficiency adds up At ChatGPT’s scale, repeated context could cost nearly $40M/year in compute alone… here’s the math

4/ AI companies are following the classic big tech playbook: locking user data in walled gardens to maximize profits More user data = more revenue 🤑 That’s why tech giants built empires on advertising fueled by their users’ info. It’s time to give back control to the users

5/ Memory will be a new moat that AI companies use to lock users in and raise switching costs🌊 We see this unfolding already with OpenAI’s new memory feature Maybe one day we’ll OAuth with “Sign in with OpenAI” 🔑 but true flexibility is still missing

6/ Lock-in will matter more as different AI models shine in areas like general, search, coding, therapy, etc. Projects like @layerlens_ai, building AI eval benchmarks on @eigenlayer, will make these differences clear and highlight the true value of portability across models 🔄

7/ Teams working on MCPs let AI models securely access & update user context from many sources but this still needs access to APIs zkTLS via @OpacityNetwork could unify data from web2 apps including those not accessible via public APIs, while keeping the underlying data private

8/ @Plaid transformed web2 fintech by unlocking financial data access and new apps Plaid for AI Memory could do the same, powering a new wave of AI-driven experiences and applications