ok tinkerers, i've got a fun toy for you
introducing baby-menu - a mac menu bar icon, and it's just a baby
it can't do anything, but you can feed it prompts and help it grow. it's what i think hyper-personalized, self-evolving software can look like
details in thread ๐
We tried to solve this at LolTravel about 8 years ago (chatting with real travel agents and rich chat UI), and the challenge is travel is not a one click buy decision, so context becomes increasingly difficult the more options that are discussed. which hotel or flight you are referring to from previously sent - even when they are sent as rich UI, it becomes a human memory problem when they are out of view.
Obviously @bchesky would have explored this years ago as well
Today is a hard day. I shared this note with the @linear team today: Weโve made the difficult decision to increase our workforce. This is not a cost-cutting exercise or a reflection of anyoneโs performance. Weโre simply reimagining every role for the agentic AI era. Weโre hiring. Weโre sorry about that.
8) AI makes code is cheap, your time is still valuable, so figuring out how to scale yourself is always going to be better than a tool that tries to scale for you.
7) You only review code when the worktree agent has reviewed their own code enough times that they (Claude) are happy with the result and submit a PR. Then you review the code, just like any other person on your team. Ask for changes and back to testing.
6) Imagine each worktree is an engineer on your team, assign work the same way (i.e. no two worktrees should be working on exactly the same feature), then simply answer their questions/help them test their changes/provide feedback.
4) I.e. trying to watch the changes and pair code with two people writing different features at the same time would be a nightmare.. and the same can be true with pair coding with a few agents in parallel.
3) Once you have a process you like, it should be the equivalent of you pair coding with a better version of yourself.
Pair coding with one person at a time is not scalable..
2) Create your own, or use plugins like compound engineering/gsd/superpowers.
Compound engineering for example can take many minutes between each prompt as it explores and thinks. It creates great output (given strong input) at the cost of time, like any person would.
1) Working on many things in parallel on the surface is very overwhelming.
So we need to start by creating a slow and intentional process for shipping high quality features (i.e. brainstorming documents, planning documents, todos, triage, multi-agent reviews, etc).