I genuinely admire the ambition here, and I do not doubt your excitement about building a working prototype. But I am struggling with the part that matters most.
You talk about a court for truth, separation of powers, and fair adjudication, but you never actually explain who decides what is true, how those decision-makers are chosen, or how their incentives are constrained. In any system that claims authority over truth, governance is not a feature. It is the entire problem.
How do you prevent this from becoming just another overlay of power that reflects the values, blind spots, and social pressures of the people running it? And how do you handle deep, good-faith disagreement where there is no clean or provable answer?
I am not trying to be cynical. I just think that building the mechanics is the easy part. Designing legitimacy, accountability, and resistance to capture is the real work.
I'm kind of bewildered that... apparently the whole thing sprung out from Claude's fingertips fully formed in one go, meanwhile I've been vibecoding something vastly simpler (and not revolutionary) for myself for 2 weeks already, and still not done. (Admittedly, I didn't have the design documents ready and I'm doing it one feature at a time.)
I admit I've not being using Claude Code out of laziness and instead doing it in a project, like a peasant. It seems that for my purposes (one feature at a time) the latest Opus and Sonnet are pretty much the same, ace the same things and stumble on the same things. Now I do kind of wonder what would happen if I gave Opus the full overview and told it to go crazy, but I suspect the result would be a surreal bug fiesta.
About the Trust thing: will give it a try when it's released to the smallfolk.
I was thinking uneasily last night about all the novel forms of chaos and confusion and harm that AI makes possible: Truth blurred by slop, minds weakened by it, new forms of entertainment addiction injected into people online. Accidental harms of all kinds made by goofy amateurs wielding tools for fools, things like Clawdbot. Criminals and governments empowered by the new technology to do supercharged spying, stealing, blackmailing, scamming, lying, tricking, persuading, hiding, punishing. Well, I thought, I suppose we will eventually develop laws and protections that curb all that some, but it’s going to be a rough transition.
It’s heartening to see someone setting out to protect us all straightaway.
A combination of things: my time got rapidly compressed during this period. So the two years was more like six months. (Had a second baby and then an increase in job duties that basically didn’t give me room to breathe.) however during that period I fleshed out the majority of what I wanted to do on paper out of my head, which is the hardest part of requirement gathering as you have to be specific. I had no ability to interact with the team though so we went on hiatus. Then Claude got really really good at turning requirements into code in December of last year. If everyone had been able to commit full time to the project I’m thinking it would have taken us about two or three months. I have no idea how to do the integration work for money so that entire piece is descoped right now. I have also descoped, for now, doing automated web searches and annotations with an AI agent because I couldn’t afford the token budget if it took off.
I genuinely admire the ambition here, and I do not doubt your excitement about building a working prototype. But I am struggling with the part that matters most.
You talk about a court for truth, separation of powers, and fair adjudication, but you never actually explain who decides what is true, how those decision-makers are chosen, or how their incentives are constrained. In any system that claims authority over truth, governance is not a feature. It is the entire problem.
How do you prevent this from becoming just another overlay of power that reflects the values, blind spots, and social pressures of the people running it? And how do you handle deep, good-faith disagreement where there is no clean or provable answer?
I am not trying to be cynical. I just think that building the mechanics is the easy part. Designing legitimacy, accountability, and resistance to capture is the real work.
You’ll see on launch day! Way more simple than you suppose!
And my prayer is, "please let some guy not be full of shit."
I also pray for this, believe it or not.
Oh I believe it. That's why I still read your stuff :)
Thank you for your goading the rest of us along too!
I'm kind of bewildered that... apparently the whole thing sprung out from Claude's fingertips fully formed in one go, meanwhile I've been vibecoding something vastly simpler (and not revolutionary) for myself for 2 weeks already, and still not done. (Admittedly, I didn't have the design documents ready and I'm doing it one feature at a time.)
Give it more context and reason why and also are you using opus 4.6 in Claude code?
I admit I've not being using Claude Code out of laziness and instead doing it in a project, like a peasant. It seems that for my purposes (one feature at a time) the latest Opus and Sonnet are pretty much the same, ace the same things and stumble on the same things. Now I do kind of wonder what would happen if I gave Opus the full overview and told it to go crazy, but I suspect the result would be a surreal bug fiesta.
About the Trust thing: will give it a try when it's released to the smallfolk.
I was thinking uneasily last night about all the novel forms of chaos and confusion and harm that AI makes possible: Truth blurred by slop, minds weakened by it, new forms of entertainment addiction injected into people online. Accidental harms of all kinds made by goofy amateurs wielding tools for fools, things like Clawdbot. Criminals and governments empowered by the new technology to do supercharged spying, stealing, blackmailing, scamming, lying, tricking, persuading, hiding, punishing. Well, I thought, I suppose we will eventually develop laws and protections that curb all that some, but it’s going to be a rough transition.
It’s heartening to see someone setting out to protect us all straightaway.
Feel free to be very very critical on the final product. I did add an AI experience (humans are the approvers) that I think might be controversial.
HUZZAH, MY LORD SOME GUY!!!!
Can you ask Claude to use common WebExtension APIs so it can deploy to Firefox from the same source code?
I can today.
So 2 years ago you were looking for a team of talented people:
https://extelligence.substack.com/p/im-putting-a-team-together
And now CC has done the thing in 15 minutes
How much of that is "your goal has gotten less ambitious" vs "you've actually done some work on it" vs "we're actually in the singularity"?
A combination of things: my time got rapidly compressed during this period. So the two years was more like six months. (Had a second baby and then an increase in job duties that basically didn’t give me room to breathe.) however during that period I fleshed out the majority of what I wanted to do on paper out of my head, which is the hardest part of requirement gathering as you have to be specific. I had no ability to interact with the team though so we went on hiatus. Then Claude got really really good at turning requirements into code in December of last year. If everyone had been able to commit full time to the project I’m thinking it would have taken us about two or three months. I have no idea how to do the integration work for money so that entire piece is descoped right now. I have also descoped, for now, doing automated web searches and annotations with an AI agent because I couldn’t afford the token budget if it took off.
AI - either our miracle of World Peace or our Terminator.