Shikhar Shukla

Building cross-platform orchestration engine for AI coding agents with CKRV.

[Shikhar Shukla @_Shikh4r_]

A couple of days ago I asked openclaw to add a few coding agents to chakravarti-cli. It took a little bit of trial and error but it managed to add 8 new coding agents. What a time to be alive.

[Cursor @cursor_ai]

We're excited to join forces with @SpaceX to advance the frontier of useful AI. Expect significant improvements to Cursor soon.

[Shikhar Shukla @_Shikh4r_]

6/ The signal for builders: inference is the migration opportunity. The training/inference bifurcation is real, and the company with the deepest moat just confirmed it with its checkbook.

[Shikhar Shukla @_Shikh4r_]

1/🧵 NVIDIA Paid $20B to Admit GPUs Aren't Good Enough

[Shikhar Shukla @_Shikh4r_]

5/ NVIDIA closed the gap through acquisition, not engineering. $20B is the price of an architectural admission. The largest AI chip startup acquisition in history.

[Shikhar Shukla @_Shikh4r_]

@shambhav15 Do more of these, its actually pretty fun to watch!

[Shikhar Shukla @_Shikh4r_]

@claudeai is down. Claude Code is still working.

[Shikhar Shukla @_Shikh4r_]

code: https://t.co/Un1uFvAz71

[Shikhar Shukla @_Shikh4r_]

A couple of days ago I asked openclaw to add a few coding agents to chakravarti-cli. It took a little bit of trial and error but it managed to add 8 new coding agents. What a time to be alive.

[Shikhar Shukla @_Shikh4r_]

@drishtadyumn Is it better than cursor?

[Shikhar Shukla @_Shikh4r_]

Great move!

[Deepinder Goyal @deepigoyal]

LAT Aerospace has acquired Sharang Shakti, an early-stage defence robotics startup based in Gurgaon.

This is our first move toward building indigenous defence capabilities alongside our long-term mission of developing next-generation civil aviation platforms from India.

Civil aviation and defence are often viewed as separate sectors. But the core technology stack is shared across autonomy, perception, sensing, navigation, guidance, and control systems.

By bringing Sharang Shakti into LAT, we are building these capabilities in-house, from first principles, with the intent to deploy them across both defence and civil programs over time.

Slowly but surely.

[Shikhar Shukla @_Shikh4r_]

@tanay_mehta Subtle flex, you can manage to run most models with about half that memory...

[Shikhar Shukla @_Shikh4r_]

@rxhit05 Nope!

[Shikhar Shukla @_Shikh4r_]

It is not perfect. But it comes very close to Opus 4.5 level performance. The code changes are mostly correct. The only difference that I can notice between GLM 5 and Opus 4.5 is that Opus is able to one shot most of the problems but GLM 5 generally gets about 90% there. I generally have to use Claude Code or OpenAI Codex to spot bugs and let GLM 5 solve them. The final code is 90-95% close to Opus level code.

[Shikhar Shukla @_Shikh4r_]

@msg Interesting project to benchmark new models. Would be fun to see how accurately Claude Sonnet 4.6 can create a similar agent visualizer. Maybe even a full Stardew Valley like village full of agents.

[Shikhar Shukla @_Shikh4r_]

@pankajkumar_dev Switch to OpenCode, use your Antigravity auth for the models. Add multiple accounts to your family group and use all of them for 6x usage as compared to a single account. Use Yazi and lazygit if you wanna see the code changes interactively in the your terminal. And use ghostty!

[Shikhar Shukla @_Shikh4r_]

Big day: Used my OpenClaw agent to autonomously code & deploy a key update to chakravarti-cli—all via text! Swapped Makefile for Justfile + fixed validation loops for buttery-smooth workflows. AI agents FTW! PR: https://t.co/z7mH1r0rnM