FYI Caffeine "general availability" schedule. Phase I — we're working to make the Caffeine master chat fully open to public, ETA 19th August. At this point, you will still need access codes to build, but code disbursement will greatly accelerate (we now have tens of thousands waiting) Note that the team has been rearchitecting Caffeine's scaling framework, for a bunch of important reasons, and when Phase I begins the master chat will already have been migrated. Phase II — we're working to give everyone access to free draft app creation, with an ETA 16th of September. All those waiting in line for access codes will receive them before general availability. Phase III — we're working to unlock pay-for-usage, which is necessary to allow people to push changes to draft app functionality into live apps at scale (live apps are permanently online and consume real cycles), and to allow people to create millions of draft apps. We're working to an ETA 7th October. To reach Phase III, we also need our new ICP blob storage service (i.e. file storage) to be online and integrated. We already have something working internally. The reason for this: our early users love creating things like their own sovereign "Google Drive", shared family photo galleries, and things like that. There's no reason to believe when we open the floodgates this will get less popular. Currently, the apps early users are creating are maintaining their files in application memory (i.e. ICP canister smart contract memory). However, that memory costs $5/GB on ICP, so once Caffeine is in production, unless the methodology that the AI uses to store e.g. media files changes, storing a 1TB photo collection would cost the user $5,000 a year, which is obviously ridiculous. The blob storage will be crazy cheap by comparison –something like $0.025/GB, which means a user will pay $25 a year in cycles for storing e.g. a 1TB photo collection. For the engineers among readers who want to know more – the new service provides *immutable* storage, which means blobs can be written once (e.g. "uploaded"), read many times (e.g. "viewed" or "served"), and then garbage collected when not needed (e.g. "deleted'), but not manipulated in place like data held in canister memory. Behind the scenes, this is quite a major lift for the Caffeine team, because blob functionality must be integrated with the way the Motoko programming language and its "Enhanced Orthogonal Persistence" system works: essentially, so that garbage collection of unneeded blobs is completely automatic, and AI can continue writing code in the same groundbreaking way on ICP that it does today – just hacking out the abstract logic, and nothing else. Phase IV — we haven't set the ETAs for this yet, but we will work towards flicking the switch that allows anyone to publish apps to the App Market in short order, so people can create cool apps through conversation, then either share them with the world for free, or resell them. I can't say much now, but Caffeine's App Market will work differently to any App Store you've probably ever imagined – and in mind-bending ways. It's going to be very cool and people around the world will gain a whole new way to make money. Phase V — we will work towards flicking the switch that allows Caffeine users to add trustless Web3 multi-chain functionality into their apps. Currently, Caffeine restricts this because of complexities, even though the Internet Computer/ICP supports this natively. For example, when you are developing an app in draft mode, it needs to interact with dummy ledgers, while live apps, which ultimately get their logic from draft apps, need to interact with real ledgers – and this kinda thing needs to work transparently, since self-writing is app development and ownership through conversation only. Phase VI+ No dates yet, but expect the following: 1) Massive acceleration using ASICs for inference 2) Enterprise features enabling multi-user control flows for pushing changes in draft apps to live apps 3) Conversational workflows that can incorporate rich media and external design documents 4) The equivalent of MCP for Caffeine, which enables rapid automated migration of existing enterprise systems 5) Hell, maybe even vibe coding hooks for craft coders (I'm messing, this for sure)...