Rohit Jain

I like building things. Software counts || Founding Engineer @kim_cc_official | BITS Pilani

[Rohit Jain @xRME__]

@obie Try this
https://t.co/es0IheASk5

[Rohit Jain @xRME__]

@scottstts @claudeai try this out, it creates a compacted copy automatically. so you have an option to start from cached session https://t.co/k5Dfd4Lcb0

[Rohit Jain @xRME__]

@scottstts @claudeai try this out, it creates a compacted copy automatically. so you have an option to start from cached session https://t.co/k5Dfd4Lcb0

[Rohit Jain @xRME__]

@theghost1623 @trq212 @thsottiaux its a replacement for chat, not output. i can collaborate with claude on many more touch points than a single append only chat allows.

[Rohit Jain @xRME__]

@trq212 I love the richer way to communicate with Claude using Claude docs, If I had to chooses one I'd choose docs over opus 5.5. It's one of the best things released in a while!!

@thsottiaux please 🙂

[Thariq @trq212]

This is the first step for artifacts becoming a much richer form of collaboration, working on a longer post and examples.

[Rohit Jain @xRME__]

@soorajchandran_ Yes, this! If I could decide whether to learn or apply.

[Rohit Jain @xRME__]

finally back to talkative claude!!

can't stand codex being all "I'll handle it" just to use do it wrong way.

tho luna 😘

[Rohit Jain @xRME__]

@Da7_Tech Hey @Da7_Tech, We have a planner executor agent system for CX where a lot of engineering time goes into implementing new execution paths, debugging failed trajectories, and fixing issues surfaced by our evals. I'll be trying swe2 on that.

[Rohit Jain @xRME__]

@Gegam245074 Yep switching back to Claude Code. Apart from Luna being absolutely awesome. Claude Code harness and their models seem to be much better in almost both cases, working with me as well as completing the task on their own.

[Rohit Jain @xRME__]

@notkevinzhang would be very useful in verification at each step in CX workflows, and even if the accuracy of the model is not much higher, having verification over each bounded function would still push the accuracy of the final response much higher.

[Rohit Jain @xRME__]

@harshil3004 Really wanted to try it out, can I get one!

[Rohit Jain @xRME__]

@reach_vb I recently switched to codex for luna and to try out astra, for any serious work I find it very irritating to use it doesn't really tell me what is doing unlike clade code desktop.
Maybe I'm not trusting it enough to let it do its thing.

[Rohit Jain @xRME__]

@MattPRD @Muse Yes, excited to try it out!

[Rohit Jain @xRME__]

@alymoursy @ycombinator @useveeza lfg!

[Rohit Jain @xRME__]

@cursor_ai Oh Thank you!

[Rohit Jain @xRME__]

@withmartian Does this analysis account for prompt caching? I’m wondering whether routing requests across multiple models reduces cache hit rates compared to using a single model.

[Narayanan (Rayn) @chalkmeout]

Appointing a 'Head of AI' is usually a sign a company is behind, not ahead. If AI is one person's department, it isn't in your product. The teams pulling away treat it like electricity: everywhere, owned by no one, nobody's single job title.

[Diana @sdianahu]

the cost of shipping code went to zero

taste didn't

but "taste" sounds mystical and unfixable, so nobody teaches it. here's the unmystical version: taste is just an eval you haven't written down yet

how you choose what to measure is what matters

1/8

[Rohit Jain @xRME__]

@mynameisyahia @getcontextdev 🙋‍♂️

[vivek @itsreallyvivek]

https://t.co/HV0irzdHax

[Rohit Jain @xRME__]

Goals

[Archie Sengupta @archiexzzz]

karpathy approved 🫡