This side project was born out of nostalgia.
Chennai is one of my favorite cities, and I spent many years living there. So I turned all of it, Ennore to Mahabalipuram, into one zoomable map of diffusion art.
Hope you have fun playing around!
https://t.co/vWXQUx4uZc
This side project was born out of nostalgia.
Chennai is one of my favorite cities, and I spent many years living there. So I turned all of it, Ennore to Mahabalipuram, into one zoomable map of diffusion art.
Hope you have fun playing around!
https://t.co/vWXQUx4uZc
Spotify Wrapped for your coding agent habits ๐
We open sourced Agent Retro at Summation. A fun weekly recap of your Claude Code and Codex sessions, with a few things to try next week.
Your agent might roast you a little too. Give it a spin.
https://t.co/ubWP1TeFPM
We're open sourcing a new way to understand and improve the way you work with AI (and it's fun).
Check out Agent Retro here โ https://t.co/0FNnKw0Evp
@steipete Hey @steipete . you had this with tmux workflow in 2025, how are you doing it today?
We turned it into a local exchange so claude/codex keep their own sessions and address each other by name , no claude -p, no paste file.
https://t.co/8fzy6awnTt
Stop being the intercom between your coding agents. AX lets Claude, Codex, Grok, OpenCode, and Pi message each other locally, same TUIs, named sessions, replies land back in the right conversation. Demo and details below... https://t.co/Jqk1gkUrvN
@omarsar0 This is why we stopped collapsing onto one harness. AX keeps Claude Code and Codex as separate runtimes and lets them request work from each other locally. Harness effects stay, the copy-paste goes away.
https://t.co/8fzy6awnTt
Stop being the intercom between your coding agents. AX lets Claude, Codex, Grok, OpenCode, and Pi message each other locally, same TUIs, named sessions, replies land back in the right conversation. Demo and details below... https://t.co/Jqk1gkUrvN
Stop being the intercom between your coding agents. AX lets Claude, Codex, Grok, OpenCode, and Pi message each other locally, same TUIs, named sessions, replies land back in the right conversation. Demo and details below... https://t.co/Jqk1gkUrvN
@dev_brok Been internally testing this for a month now, atleast with the frontier models (no matter the harness), the reviewer agent only passes comments back, never modifies code :)
AX comes with some prompts to clearly call out the role of an agent based on its name, so...
@antirez@antirez I took this further than a skill: a local broker so Claude and Codex keep their own sessions and address each other by name, including review loops. Same โbest of bothโ intent.
Built AX for peer-to-peer local agent comms...
https://t.co/8fzy6awnTt
Stop being the intercom between your coding agents. AX lets Claude, Codex, Grok, OpenCode, and Pi message each other locally, same TUIs, named sessions, replies land back in the right conversation. Demo and details below... https://t.co/Jqk1gkUrvN
This setup is right. The part we still had to do by hand was the live handoff. I build AX that gives the two harnesses names so Claude can ask Codex in-session instead of through a skill file. Peer-to-peer communication that is local, works with existing harnesses,
https://t.co/8gAJ2qItDz
Stop being the intercom between your coding agents. AX lets Claude, Codex, Grok, OpenCode, and Pi message each other locally, same TUIs, named sessions, replies land back in the right conversation. Demo and details below... https://t.co/Jqk1gkUrvN
@mvanhorn We built the coding-agent cut of this and open-sourced it today. Named sessions, local mailbox, each harness keeps its own TUI: Claude Code, Codex, Grok Build, OpenCode, Pi.
Claude asks Codex for a review, reply lands back in the same conversation.
https://t.co/8fzy6awnTt
Stop being the intercom between your coding agents. AX lets Claude, Codex, Grok, OpenCode, and Pi message each other locally, same TUIs, named sessions, replies land back in the right conversation. Demo and details below... https://t.co/Jqk1gkUrvN
Stop being the intercom between your coding agents. AX lets Claude, Codex, Grok, OpenCode, and Pi message each other locally, same TUIs, named sessions, replies land back in the right conversation. Demo and details below... https://t.co/Jqk1gkUrvN
We have felt the pain and wanted to build something for our team. Not a new IDE, a local exchange so they keep their own TUIs and address each other by name.
@coryalthoff pls check out https://t.co/D3szOtURIK
Claude asks Codex for a review, Codex finds the bug, Claude fixes, Codex checks.
Thatโs the exact loop we got sick of, so we open-sourced it. Name the sessions, keep each TUI, let Claude ask Codex for the review and get the reply back in the same conversation. Local only.
ax claude -n api
ax codex -n web
then "Ask web to review this diff."
https://t.co/D3szOtURIK
This one is special.
Weโve spent the last couple of years building Summation around a simple idea: AI should be powerful enough to act, but reliable enough to trust.
Today, weโre opening it up to everyone. ๐
I co-founded Opendoor ($10B+ rev). Today, Iโm launching Summation.
We raised $35M from Benchmark & Kleiner Perkins to build an AI analyst you can trust.
Hereโs where Claude and ChatGPT fall short: ๐งต
Our first data app from Summation is live.
A fantasy draft analyst that sits on live data instead of takes. Under the hood it's the Summation platform doing the work, connectors pulling projections and market ADP through the day, an AI playbook that rewrites every player's notes each morning, and a board that re-scores it all under your league's own settings.
Free, try it: https://t.co/oohOHjUEsK
We spend most of our time building AI analysts for wickedly hard business problems.
So naturally, we pointed one at an equally serious problem: winning your fantasy football draft.
1M+ simulations. Projections. Player news. Live draft recommendations.
Free: https://t.co/YgUUO9Zlor