Today i came across someone who said :

One thing I absolutely love about AI tooling, specifically Claude Code, is the ability to be building from anywhere on my phone, e.g. the dog park.

Needless to say, Darwin loves it too!

Where do you work these days?

“The dog got a lot of attention and love” …. I am not sure how/what to feel about this. The question is : Is it production/important work OR a toy project…. ( I know the answer to this .. ) Out of curiosity how many here do the same.

I gave it a try some time ago (my setup was using OpenCode + self-hosted model on DGX spark) Mainly was maintaining an OSS project (pandera), fixing bugs, adding features But it got old after a few weeks… at some point I still like focused work on my work desk.

I presume you have an extensive test suite, and

pandera

was build before LLMs. I think this will be moderately ok if you have extensive test suite. I once thought about using openclaw and use it as such, but never got the courage to execute it. I am always scared it will do this that i am not aware/don’t understand and all i will be left with is “prompt” it when something goes wrong. I totally agree, too much technology is detrimental to human mind.

yeah, with Pandera it has a well-established API, linting, type hint checking, unit tests, etc. So in that case my agents had a lot of context and conventions to work with

Hey @broad-monitor-993, Could you try the below command in your DGX spark:

cat /proc/sys/vm/mmap_rnd_bits

Mine shows the bits as

33

which i think is custom for this machine. Does flyte buildx container (part of devbox install) able to create multi platform builds on your spark machine. You can try to build a container with ImageSpecs (paltform =“linux/amd64”) I presume you have GB10 from Nvidia. Mine is from MSI. I want to confirm the OS settings are the same