@mikker

431 Followers
288 Following
845 Posts
Mikkel Malmberg ✹ solo founder https://10er.com and more ✹ doing YouTube at https://youtube.com/@mikkelmalmberg ✹ building https://nitrokit.dev ✹ maximum overbusiness as https://brainbow.studio ✹ @mikker most places ✹ I have three kids ✹ 📍 Denmark
Websitehttps://mikkelmalmberg.com
Twitterhttps://twitter.com/mikker
GitHubhttps://github.com/mikker
Blueskyhttps://bsky.app/profile/mikker.dev
mikkerMar 15

I love how agents will be hesitant, inheriting caution from their feeble creators.

"This is a big refactor, it might take a while. Maybe just do the small version now."

Absolutely not, my boy. Go, run like a wild horse. I'll see you in 5 minutes when you're done.

Show threadmikkerMar 13
Yesterday I had Codex write all the first-pass docs for Tuna and it got it 90% great. I edited 5% and decided to not care about the last 5%.
Show threadmikkerMar 13

Now that I don’t have to care about most of it on my own stuff, it’s suddenly staring me in the face how much faster one can move, when you don’t hem and haw over every single micro decision.

And now I’m wondering: how could this look for every other aspect of my business? What if I didn’t care so much about the tone of the marketing. What if I didn’t care so much about the shape of customer service. What happens if i make it about the outcome and not the structure?

mikkerMar 13

It is increasingly obvious to me how much energy I used to spend on making the code a product on its own. So much thought into the right abstractions, the right structure. Not the outcome, the structure.

The goal of this was to have a codebase that was easy to work in, easy to reason about. Easy to maintain. Almost none of it was to further the product. If so, only in a sort of meta work indirect way.

mikkerMar 12
Considering enforcing a strict release train with feature freeze and all to keep my split git worktree personalities to align. As soon as one is about ready to release, another has started a new epic
mikkerMar 12
actually, having a folder full of markdown files written by and for claude to "maximize productivity" is the exact opposite of productivity
mikkerMar 9
mikkerMar 7

The Claude frontend skill outputs are already recognizable. Off-the-shelf aesthetics become cringe so fast. See Bootstrap, Material, etc etc.

There’s no way around learning what goes into good design. That means open eyes, lots of practice. Easier than ever but fundamentals same

Show threadmikkerMar 6
This lets me, the reviewer, have a chance of seeing what you can do with AND without AI. And I don't have to squint all the time to try and guess whether or not you added Tailwind 3 instead of 4 or the bots did and you didn't care.
Show threadmikkerMar 6

Then take a few rounds of AI assistance AND add a note in the commit messages. "(Codex assisted)", something like that.

Do all the things you didn't have time to. Fix that one bug or edge case. Add the tests. Make the README pretty. Basically anything that shows good effort, taste and luxury.