We call it Real-World AI: software that takes real actions on your behalf, with your approval, instead of only producing text in a chat window. That's the whole idea in one sentence, and it's worth stating plainly up front because it's a genuine shift, not a rebrand. Dropstone started as a coding agent. It still is one. But over the past several months it also learned to draft your email, place real phone calls, take calls back, watch your inbox while you're away, and turn the lights off in your living room. None of that was the plan when we started. All of it followed from the same underlying bet. This post is the story of that shift: where it started, what we built, what we're calling it, and where it goes next. Dropstone began as a coding agent, and a fairly narrow one at that. The pitch was simple: frontier-class coding intelligence, running on the best open-weight model each cycle, priced like infrastructure instead of a luxury good. That product still exists. It's most of what Dropstone 1.5 and 1.6 are about. Coding was the right place to start because it's a world made entirely of text. A file, a diff, a terminal command: all text in, text out. That's exactly the environment language models are best at, which is why AI got good at coding years before it got good at almost anything else. It was the easy end of the problem. The harder end of the problem is everything that isn't text: an inbox that needs a reply, a business that needs a phone call, a thermostat that needs adjusting. Those are real-world actions with real-world consequences, and getting an AI system to touch them safely is a different problem than getting it to write good code. We started with email and calendar, because the stakes were manageable and the value was obvious. Dropstone reads your inbox and drafts replies and calendar events. It does not send anything on its own. Every outbound email sits behind an approval you have to click. That approval gate is not a limitation we plan to remove later. It's the actual design: the model drafts, you decide. Email was the safe first step because a draft sitting in an approval queue can't do any damage on its own. A phone call is a different kind of commitment: once it's placed, it's happening in real time, in the real world, with a real person on the other end. Dropstone places real phone calls on your behalf. Book the appointment, chase the refund, ask the question you've been putting off. And it works in both directions: when the person you called calls back, Dropstone can take that call too, and pick the conversation back up where it left off. That's a meaningfully different claim than "can dial a number." It's closer to having someone who actually handles the call, start to finish, and tells you what happened. Everything up to this point still requires you to ask. The next step was making Dropstone useful when you don't. Proactive monitoring watches your inbox in the background and surfaces a briefing, even while you're away. It's opt-in, and it's narrow by design: unread mail from the last few days, nothing invasive, nothing that reads like surveillance. The goal isn't an AI that watches everything. It's an AI that notices the handful of things that actually needed your attention today, so you're not the one doing the noticing. The most recent surface is the one furthest from a keyboard: the physical devices in your home. Through connectors like SmartThings, Dropstone can see and control TVs, thermostats, lights, locks, speakers, and a growing list of other device types, reading live state and acting on rules you set up yourself. This is also the surface where we're being the most careful. Smart home control has existed for a decade through Alexa and Google Home; the interesting part isn't that an AI can flip a switch, it's that the same agent that drafted your email and made your call can also turn off the lights, using one continuous memory instead of three disconnected apps. We call this Real-World AI: an assistant that acts, not one that only answers. Like Software as a Service, it's a category, not a product name, and we're not claiming to be first or only. It's the name we're using for the pattern we've built toward over the past several releases, and it describes a shift the whole market will build toward, not just Dropstone. Email, calls, monitoring, and smart home devices are the surfaces that exist today. They're not the ceiling. The same runtime and the same approval-gated design that made those four safe to ship is the thing we're extending outward, toward deeper hardware integration beyond what a connector like SmartThings can reach on its own. We're not going to attach a date or a device list to that yet, because we'd rather ship it than promise it. What we can commit to is the shape of how it'll arrive: one agent, one memory, expanding the number of real-world surfaces it can act on, with your approval required on anything that matters. Dropstone is live today at https://dropstone.io, and the Assistant is at https://chat.dropstone.io. If you want the technical side of how the coding agent underneath all of this is built, start with our Dropstone 1.6 post, or the full technical report at https://blankline.org/research/dropstone-1-6-technical-report.