20% that's my estimate
Whenever I start to really really spend lots of time directly with Claude, it can only mean one of two things - My systems, validations, context, prompting have weaknesses, and I can't launch Claude to run independently
- I'm taking on hard/novel problems, and bringing in insight, discovery, creativity into the mix
Pretty happy to be at 20%
@VictorTaelin Thereโs always this hope. If I just update the CLAUDE.md file, if I just add one more test, if I just add another /validate skill, if I just uploaded the entire spec, if I justโฆ.
@krishnanrohit So much of the software that would have been uneconomical to write can be written now (test automation, one off apps etc)
They all feel like progress, but we used to make progress without them.
It reminds me of this - https://t.co/Ltwp6ere5U
@tszzl This makes me wonder. If the question I really want to ask isnโt are they conscious, but do they have a desire for revenge and can they act on that.
@theo Can you solve permissions? I would like some people to have access to some folders, and other people access to other folders/files.
Also, imagine a csv, where you can only update the rows, where userid == $USERID. Or even read the csv, and get a different version
Earlier this year I was getting frustrated with Claude's charts, fed this book to claude and had it generate a Tufte skill. Instantly got simpler/more beautiful visualizations.
https://t.co/lfXwyQfmQG
I've been using AI actively to help me manage community volunteers and events. What has worked, and what hasn't
Works:
* Using AI for follow-up - All the reminders, knowing when people responded, knowing what's delayed, following up with every "interested" volunteer - It's been great for keeping things from dropping
* Using AI for alignment - For the committee, what did we discuss, actions items, open questions
* Using AI for operationalizing - After the event/program - managing retros, and coming up with operational checklists for the future, is a place that AI really shines. It needs a ton of context to do this right, but when it has it, it works wonders.
Doesnt:
* Sharing institutional knowledge is hard. I can't partition and share my personal git repo. Organizations don't have a "document store" for volunteers. Others aren't using AI much yet, but if they wanted to, they would have to start from scratch
* Some channels are off-limits. Can't take voice recordings in meetings, too invasive. WhatsApp automation is hard/scary. Keeping things in sync when there are "black holes" is hard.
* Making everything a spreadsheet doesn't work for volunteers. They have their own agendas, personalities, limitations. Making everything into a form, and forgetting that you are privileged to have their time, leads to poor engagement. Don't AI the personal touch.