no. 001 — immovable object
an AI that reads your chats back to you. the surveillance machine, pointed inward.
----------------------------------------playing offense — the founding essay
We all know the surveillance and profiling issue — it is not new and everybody is aware of it. Everyone who has at least once opened Instagram can tell how the doomscrolling algorithms morph our feed live to maximize retention and stuff us with ever more impressions. With the onset of AI this question has taken a new twist: mass surveilling and dopamine-maxxing with AI slop have become the new reality of social media and the web. In this light, many people have been raising concerns about where it all heads, and the corporate leaders' usual answer was: “Well, yes, there will be an ever stronger AI to snipe you, but there will also be an ever stronger AI to protect you.” And while I believe that this certainly happened, and we are indeed now surrounded by all sorts of AI security features, I think I have a conceptual disagreement with the current implementation of AI protection.
I think that privacy now is playing defense. I will explain what I mean. Sniping is proactive — on every page on the internet we are being actively profiled. We are not kindly asked to take part in a poll or to rate the content (this also happens, but the product designer in charge of this kind of thing won't last long in their position). Neither are we in charge of our feeds and suggestions. Instead, the platforms we are using are simultaneously a) a highly sophisticated surveillance machine and b) a proactive decision maker that steers our web experience. So the tooling it uses is twofold proactive: it proactively profiles us and proactively shapes our web.
On the other hand we have the privacy advocates. What are the tools that are on their side? Well, we have mainly three: first, “trust me bro” corporate-exec promises; second, the acknowledgement option, where we are allowed to limit the surveillance to the limited only; third, anti-technology technologies that will make a brick out of your phone. That is at least what I am personally aware of. The tools listed are at most very defensive and at least just ridiculous. What I think is that they should play offense.
What does it mean to play offense? 2026 has become the year when the goods promised by AI became so tempting that we were ready to sacrifice our privacy and ownership. “The age of the Lobster,” marked by the release of open Claw, showed how, given absolute freedom, AI can be extremely effective. Everything kind of converged at one point in time — 2025's agentic hype, 1M-token context windows and general reasoning capabilities, MCPs, etc. For many, Claw was the moment when it just clicked: it does things autonomously, and I TALK TO IT via WhatsApp.
For me the moment when things clicked happened at approximately the same time, but differently. I was working on some project with Claude Code and somehow found myself in the .claude folder (where, as you can understand by the dot in the beginning, I was not supposed to be). There I saw the memory.md file — a little profile of me that Claude kept to increase the conversation quality. Opening it, I was amazed by two things: the deep portrait of me that Claude had created, and the fact that it was never exposed to me.
The former impressed me with how it focused on the implicit — it was not a combination of my previous prompts. It was a critical summary, highlighting my psychology. None of what I saw there was actually written or explicitly said by me — it was all analysis done by Claude.
The latter was rather troublesome — I felt like I was being spied on. I prompted Claude, saying: “Don't you think that things like this should be somewhat exposed to the user and not hidden in a hidden folder?” It would a) be ethical and b) be extremely insightful. And while I was aware that ethical issues are not something to discuss with a tool coming from Silicon Valley, the second part of that was something that stuck in my mind. I thought that this was nothing but an impartial, data-informed portrait.
The first intuition I had was: I want to dump all my chats into it. I want it to think through all my relationships, to analyze my behaviour, to see the patterns. This was something primitive and very direct. Then I started to think about Claw and how funny it was for people to talk to a machine. It led me to the biggest game changer about the AI tool: verbal communication with machines. I started thinking — we made machines listen to us, but what I want is to make a machine talk to me.
I see that it knows a lot about me. I open the network tab and I see hundreds of requests made on every link I click. I see tons of cookies hung on my browser. I see how Instagram profiles me — I once showed my girlfriend that I can turn my feed into PornHub in 3 minutes. I was scrolling at the speed of light until I saw a pretty girl's face. I deliberately paused on it. Then scrolling again until the next pretty face — this time wearing a bikini. A couple more iterations and I have only OnlyFans creators doing the nastiest things Instagram can technically allow. I saw how the machine learned about me (although it was tricked into the specific assumption that I was a gooner) — but I wanted to know exactly what it knows about me. The same happened with me googling something and seeing a suggestion regarding it on YouTube the day after. At this point I thought: I know how to talk to a machine, now I want to make the machine talk back.
At this point I recalled the old philosophical dilemma: “What happens when an unstoppable force meets an immovable object?” Profiling tools are so good because a) they are proactive and b) they have all data imaginable available to them. They play attack, and I wanted to have someone by my side who plays attack too.
I want a machine to keep a close watch on things I am not aware of. I want it to analyze my traffic. I want it to reverse-engineer algos. I want it to analyze cookies and tell me what's going on. I want it to read my messages and tell me to cool down when it sees me repeating the same pattern and going aggressive in a chat again.
I already have some ideas of what this means in terms of solutions. But I will keep this essay open-ended. We are yet to discuss many technicalities, and what is possible and what is meaningful. But here I wanted to share the spirit of what I want to build.
----------------------------------------← @synd inc. · no cookies. no analytics.