That’s a really sharp observation. Hopefully they take a belt and suspenders approach to these smoking guns in future.

One shot just means the user doesn’t have to iterate on it via the agent. The agent does what ever it needs to deliver the best outcome, including its own running and iteration until it’s happy with it. This could be a short or long process potentially depending on the task.

There are plenty of examples of law makers passing policy to make online and digital surveillance easier.

See TOLA in Australia, the UK trying to backdoor iCloud, Lawful Access to Encrypted Data act.

Surveillance is often sold with safety as the primary narrative.

Could you say more about your workflow? I don’t think I’ve ever gotten close to an hour of thinking before. Always curious to learn how to get more out of agents.

I don't think it's something special about my workflow and more the application area--I'm writing a lot of Lean lately and particularly knotty proofs can take quite a lot of time. Long thinking intervals are more of a bug than a feature IMO: Even if Claude can one-shot the proof in 40-60 minutes I'd rather have a partial proof in 15 and fill in the gaps myself.

This already exists for me on iOS? Maybe check your notification settings?

I’ve tried everything. Regular Claude chats notify me fine, but nothing from Code - neither a cloud session or remote control.

I had to setup a hook with terminal-notifier to get it to work myself

The models are the same. The agent implementation is different. I can confirm Claude Code performs much better than GH Copilot with the same Claude models.

Can you explain your use case? I’d be interested to understand.

Every legitimate use case for AI. It is a way to mark legitimate work done using AI tools as inferior.

This might be acceptable if it prevented or limited nefarious use cases. But it does no such thing. It doesn't help at all on that front actually and is not a problem that can be solved by technology alone.

I view SynthId as more of a method of control. It's a way for Google to label work produced by an individual using their tools as their own.

I much prefer open models that let me be creative, write code, etc.. without trying to control/track/mark me.

Same goes for Claude Code. Literally has vim bindings for editing prompts if you want them.

Codex has Ctrl-G to start an $EDITOR of your choice, FWIW.

CC is the clunkiest PoS software I've ever used in terminal; feels like it was vibe coded and anthroshit doesn't give a shit

All of these agentic UIs are vibe coded. They advertise the percent of AI written code in the tool.

which begs the question: which came first—agentic AI tools or the AI that vibe coded them?

…the AI…

The list is made by the folks who maintain the “build your own x” repo. This is pretty close to their domain.

I usually just type with two thumbs and can type pretty quickly. Swiping always felt a bit awkward to me because my phone is too large to use one handed with one thumb swiping, and swiping with a finger felt awkward compared to just holding my phone in both hands and typing with both thumbs.

I imagine if you look at how most young people use their phone, it will mostly be the two thumb method and they will likely be very quick with it.

Hm. Hadn't thought about size. I always make sure to get a phone that I can use one-handed, but I guess big phones are far more popular.