prateek

people don’t understand this take cause they don’t understand what’s happening in AI memory. Everything is moving to git backed files accessible via grep-type-systems or semantic plus grep which isn’t very defensible to offer as a service. In other words… the SOTA approaches to memory are now just agent plus terminal. And all the fancy approaches like knowledge graphs are getting rekt by an agent plus a terminal. Your fancy agent structure is getting rekt by a model that can keep track of anything over 1000+ terminal calls.

this bank cheque was made with GPT Image 2 detecting document fraud is about to get exponentially harder the big labs will try to prevent this with watermarks and refusals, but once this capability hits open source models in 6-12 months, it'll be impossible to restrict some approaches we've seen to detect this: - visual artifact detection (noise, lighting, pixels) - metadata + device fingerprints - cross-validation against well known templates none are silver bullets, and this will be a cat-and-mouse game forever we work with some of the biggest fintechs, lenders, etc and every single one of them relies on documents as proof of something: - income (paystubs, bank statements) - address (utility bills) - loss (invoices) - identity (IDs) that entire assumption just got a lot weaker, and companies will need to adapt VERY fast

hot take: tooling and infra are the only product spaces left

tui coding agents are done for. the new wave will be of html explainers for the architecture, codebase, functions and of course for the newly generated code. you'll be coding in the browser mainly. maybe some sort of electron/tauri app. native apps are going to struggle. it's really expensive to make generative frontends in swift and so on due to the compilation steps. html will win the interface wars.