Your job as a tech leader now is to get claude code max, default to opus, get your local dev env setup agent friendly, then do the same for everyone on your team. Learn how to be super human and then teach everyone else.
@jespr That’s a good one. Honestly AI naming my branches is the best. I just change from what they pick when I get a great idea out of nowhere. The rest of the time my branch names are akin to Steve Jobs wardrobe.
Went back on Code with Jason.
We got into how I ended up buying Fireside, financing it with home equity, why support is the product, and the scrappy APM I'm building.
https://t.co/4ql8PIQpdo
💡Recent insight: gaslighting @claudeai seems to improve code quality >90% of the time.
“You overengineered this, there is a simpler way”
“There is a smaller delta that buys us most of the benefits”
“There is a more elegant way”
“This is not architecturally coherent”
…before I even read its code. 😆
I’ve tried vanilla CC, conductor, superset, superconductor, Claude mac, codex mac/cli, cmux, and several others. I feel like all of them still miss on focus and autonomy.
Even in this agent world I want to focus on a few things at a time at most. Hide the rest. Pins help but only if you can hide the rest and I haven’t found any that let me.
For autonomy I’d like to see more of them make it easier to loop and run things that work when I’m gone. Idk how or what I want but I know I want it.
AI/Conductor update: I have ~30 workspaces at any given time. I only actively work on 2-5 and use pins to track top priorities (5-10).
I try to keep repos flipped closed and focus on pins. Wish I could actually entirely hide repo list most of the time AND that it was easier to create a new workspace for any repo and insta pin it as that is my flow now.
Performance has been pretty fine using it in this way.
I turn steer mode on. That was one of the last things that I missed in vanilla claude code.
Edit mode makes it easy to see whole file which was main reason I still opened an editor on occasion.
I use gstack for anything substantial (/office-hours, /plan-eng-review, /qa and the agent browser are my fav).
Only using superset for quick updates to main where I don't want all the ceremony of review, pull request, etc. that conductor forces.
I have a 20x claude and 5x codex. I use claude opus 4.7 as main driver. codex for review. Sometimes codex for review fixes and for any problem claude is having an issue with (rare).
I'm spending more time finding ways to get more verification before things need my attention. I do a lot more simplify this code and explain all the network calls to me and how they can be reduced/improved before I really look at anything. But I still look at data model hard, general interactions lightly and don't look at UI code. 🤣
@TimHaines Sounds like it. We’ll see. If so I guess I’ll switch to superset which just fires up a terminal or conductor will do that. I like terminal Claude code fine I just use conductor for the full flow and everything in one place.
Happy to announce that I've finally hit my stride on living a balanced life with AI. 🤣
For the past month I've pretty successfully stopped prompting into the evening and restored some relaxation.
Ok folks. I know I've been all in on Claude for half a year, but Codex has jumped back into the mix. I started by using it for reviews and now I'm starting to actively compare it with the same prompt to claude. Feels faster than it use to be.
@strzibnyj Hasn’t clicked with me yet. Read the book. Still feels like I’m giving up some platform protection like security and stuff. What do you use to load balance across hosting platforms then? Cloudflare?
Resiliency is always a trade off. Deploy to two platforms so you can be available when one goes down? Now config, deploys, metrics, everything is split. You need tooling to join or extra cognitive load.