A few days ago I built my own personal AI assistant. I got tired of the context switching between Claude and ChatGPT, of conversations that lose my context, and of memories that grab random details that add nothing. It's still super early and super basic, but it already helps me a lot to write important messages, because it has all my context in one place, I taught it how I like to write, and it keeps improving itself with my feedback. The funny part is that I barely did anything to create it. I told the AI what I wanted, and it created its own skills and a knowledge folder where I keep loading my info. I was telling my brother about it at 3am, super excited, the same night I made it.
Where I see the difference the most is in my meetings. I started recording all my calls and saving the transcripts, trying to get everything into one place (a16z just published that most work conversations are now being recorded by default, which is basically the same thesis I had when I started doing it). The summaries the recording apps give me are bad, because the model behind them doesn't know who we are, our stage, which clients we sell to, or anything about our roadmap, so it summarizes a meeting between strangers. When I load the same transcript into my assistant, I get a real summary with actionables custom to what we are actually building. And the whole difference is the context, because the models behind are the same ones.
But if I look at what I am doing, I am carrying my own data by hand into the one place where the AI can see it. That is dumb manual work that an AI should be doing for me, and it only exists because today our data lives scattered across different places. The calls live in one app, the chats in another one, the emails, the docs, the notes, every piece lives somewhere else, and none of these models can reach all of it. It will feel so weird in a few years to remember that these tools were this powerful and still couldn't see our own data. Because this will get solved, and very soon. Maybe next year something comes out that takes your whole computer, runs a thousand agents in the background for a full day, and comes back with your brain. Creating your own AI brain will be one click.
When that happens, the way we use AI will change shape again, so I want to define now the term I think comes next. First we had LLM calls. Then we had agents, which are many LLM calls chained with tools in a loop. For me the next one is "tasks", where you hand over an outcome and the system launches many agents that work together, and even run other agents, until the thing is done. The founder of OpenClaw wrote a few days ago that you shouldn't be prompting coding agents anymore, you should be designing loops that prompt your agents, and that is exactly it. I have been thinking about this for a long time, and it is the closest description I have seen of where all of this is going.
The limitation for these tasks today is the same one from before. None of these AIs has the perfect context of you, and even the newest tools like OpenClaw are still very early on this. In that same chat, my brother told me it is hard for the AI to replicate his brain because it lacks too much context, and that is exactly the direction all of this is heading. Once the data problem gets solved, these systems will start replicating your brain, not only for memory, but to learn from what you did, your own patterns, your way of writing, how you act in each situation. And the people that have been actively working with computers these past years, sending their messages and instructions from there, will have a much better time than the ones that haven't, because all that history is exactly what the AI will learn from.
So if the assistant will have all my context, and tasks will run entire days of work on their own, the question becomes what is left for us. My answer is that execution will become a commodity, just like coding already did. You will have your personal assistant with all your context and your computer, it will most likely live in the cloud so you can manage it from your phone, and you will deploy plenty of agents from there. And when everyone can execute like that, the value will be in the long term thinking with the right mindset, in knowing what is worth executing and where everything is heading.
We are very early, and all this context switching we do today doesn't scale, even less in a world with AI that should be the one solving it. So I think this will happen in the next months, not years, and when it does, people's lives will become 5x easier from one day to the next. Like everything I write here, this is a thesis and my next years are the experiment, but by the pace everything is advancing, I don't think I will have to wait long for this conclusion.
- my AI assistant (one-shotted, 0 changes done)