Then for extra fun, I implemented my own browser from scratch, trying to replicate their experiment, but with only one agent. 20K LOC, written about it here: emsh.cat/one-human-on...
Then for extra fun, I implemented my own browser from scratch, trying to replicate their experiment, but with only one agent. 20K LOC, written about it here: emsh.cat/one-human-on...
> Build a browser based on this code
but more like:
> I want to build a browser, lets start with just plain text 'hello world', you can look at the source code at /blah/blah for inspiration"
for example.
Hopefully this helps a bit :)
> Build a browser based on this code
but more like:
> I want to build a browser, lets start with just plain text 'hello world', you can look at the source code at /blah/blah for inspiration"
for example.
Hopefully this helps a bit :)
Second commit, only add --screenshot = github.com/embedding-sh...
And so on, the smaller pieces you can do, the easier it'll be for the agent to do it properly.
Second commit, only add --screenshot = github.com/embedding-sh...
And so on, the smaller pieces you can do, the easier it'll be for the agent to do it properly.
I think you might be trying to give it too large chunks. Try to "divide and conquer", make each step as small as possible.
This is the first "real" commit of the browser project: github.com/embedding-sh...
I think you might be trying to give it too large chunks. Try to "divide and conquer", make each step as small as possible.
This is the first "real" commit of the browser project: github.com/embedding-sh...
I think they're just dumb, there is no QA departments left, no one cares about quality, and all the people with "Good Taste" (emsh.cat/good-taste/) have moved on.
I think they're just dumb, there is no QA departments left, no one cares about quality, and all the people with "Good Taste" (emsh.cat/good-taste/) have moved on.
Bit like claiming Linux can run Windows binaries.
> Yeah, but just install VirtualBox and run a Windows VM, then your Linux can run Windows binaries!
If you need a VM to do it, then the OS does not support it :)
Bit like claiming Linux can run Windows binaries.
> Yeah, but just install VirtualBox and run a Windows VM, then your Linux can run Windows binaries!
If you need a VM to do it, then the OS does not support it :)
Worst case? Something has the wrong name.
Worst case? Something has the wrong name.
> Drag and drop son, drag and drop
> But isn't that still hard today in 2068?
> Yes son, it is and will always be
> Drag and drop son, drag and drop
> But isn't that still hard today in 2068?
> Yes son, it is and will always be
news.ycombinator.com/item?id=4678...
news.ycombinator.com/item?id=4678...
I'm using Codex, tried Claude Code and some others but everything seems worse. I pay ~200 EUR/month for the Pro plan, never hit any limits AFAIK, prorating days I guess it'll be like ~19 EUR for three days. Model used for everything was gpt-5.2 with reasoning effort set to xhigh.
I'm using Codex, tried Claude Code and some others but everything seems worse. I pay ~200 EUR/month for the Pro plan, never hit any limits AFAIK, prorating days I guess it'll be like ~19 EUR for three days. Model used for everything was gpt-5.2 with reasoning effort set to xhigh.
Just one agent (codex) + me, side by side in the terminal, me taking notes on the right, codex doing it's thing on the left :)
Just one agent (codex) + me, side by side in the terminal, me taking notes on the right, codex doing it's thing on the left :)
Generally consider myself a generalist :) Through the years been focusing a lot on testing, infrastructure and architecture/design, guess that's helpful now.
Generally consider myself a generalist :) Through the years been focusing a lot on testing, infrastructure and architecture/design, guess that's helpful now.