@DanielSmidstrup I have only tried Codex 20/mo and it can barely get useful work done compared to what I got out of Claude @ 20/mo. But I cant compare head to head.
I was a teenager. The nightmare scenario was when one of your friends bookmarked American Nihilist Underground Society (https://t.co/qN7M4GLAoh) on the family computer.
You had a lot of explaining to do if you were to use the computer the next month. It lived rent-free in my mind more than fun sites like Geocities.
Anthropic engineer:
"You're doing in 30 minutes what takes one shortcut to do in 5 seconds. You just don't know it exists."
Then she showed the exact prompting playbook that makes this possible.
Commands, shortcuts, and patterns most people never discover on their own.
Watch the video, then save the cheat sheet below 👇
Mercury is the best small business banking experience I have ever experienced. I can invoice customers from it, its got an amazing MCP that I can use to pull all the accounting stuff right into my agent and it can do my books for me. Highly recommend for small startups that want an edge.
My workflow for this has been to merge all PRs into a release branch and then I do testing there.
The simplest way to do this is just to use zellij, open a worker window and tell Claude to use it split out features into worktrees. It’s a poor man’s orchestrator but works really well.
@trq212 There are SO MANY times where I wish I could talk to my terminal like a co-worker nowadays. I really wish that the mobile app "talk to claude" would be in terminal. I tried emulating it with MacOS's `say` command but its terrible.
This is amazing. I'm noticing particularly with Opus 4.8 that it feels "talkier" than previous models.. I've since switched to HTML decks for understanding code changes, concepts in planning mode. Its much richer than markdown and lets me share and convey ideas better, especially when its wide-ranging, architectural changes where you want team feedback.
I really think this prompt will help the dialogue though! The thing that is so spooky is that as the models are getting more capable, they are making more wide-ranging decisions at each turn. The code-review bog down isn't that much of a deal for me, its more planning that is starting to hurt for me right now after I automated the reviews (its much more thorough than me).
been asking others at Anthropic how they stay in the loop with Claude and fully understand the work being done
this is one of my favorites from Suzanne:
@mattwensing Idk - mileage does vary. You gotta invoke your heroes. It’s amazing when you tell it to channel their inner Elena Verna, Meng Tu, or Gary Bernhardt. All of the sudden it comes to life in every discipline.