Pool launches AI screenshot memory app as personal assistant test

Pool is a small product, but it points toward a bigger shift: personal assistants that are built less on generic web data and more on the messy, visual traces of individual lives. By treating screenshots as a first‑class data source, the app is effectively constructing an ad‑hoc, user‑centric knowledge graph—recipes, receipts, trip ideas, product research—that future agents can reason over. That’s materially different from today’s chatbots, which mostly see whatever you paste into a single session. ([stechtimes.com](https://stechtimes.com/en/article/pool-turns-screenshots-into-a-searchable-ai-memory-layer-mq9t7na1))

For AGI, the implication is that memory architecture and data provenance at the edge may be as important as gains in model size. Assistants that live inside camera rolls, not just email and documents, will gain a much richer model of user intent, routines and preferences. If startups like Pool can solve privacy and reliability, they create a path for labs to train and fine‑tune on highly personalized, multimodal streams at scale—exactly the kind of data that can make agentic systems feel uncannily competent. The risk, of course, is surveillance: screenshot‑level access is extraordinarily intimate. How the industry handles consent, on‑device processing and data retention in these early products will set norms that either constrain or supercharge the next generation of personal AGI‑style assistants.