RC

Cofounder and CTO @_summation. Decision grade AI for enterprises.

Prev @opendoor, @thoughtworks

[RC @rcdexta]

Same dataset. Same prompt.

Claude generated a Word doc.

@_summation generated a living, collaborative artifact with traceable numbers and verifiable insights.

AI is moving from “content generation” to “decision infrastructure.”

Head to https://t.co/jbEHRi9Gsb to know more!

[Summation @_summation]

Same dataset. Same prompt: run an MBR.

Claude generated a Word doc.

@_summation built a living artifact — every number traceable, every insight verified, built for teams to collaborate.

→ Try it yourself: https://t.co/mDG5EMYwVd

[Summation @_summation]

Most GTM leaders make the call when it's too late.

A weekly GTM review needs:
1/ Pipeline coverage + PipeGen
2/ Rep activity + attainment
3/ Where leadership should intervene

@_summation generates the readout, so leaders spend Monday acting, not stitching.

👉Weekly GTM readout: https://t.co/kgjAXUBy6g

[RC @rcdexta]

This is a sharp idea: turn your edits, preferences, and “that’s not quite right” moments into reusable agent training signal

https://t.co/SC1YNvei34

[Rachid @rachbelaid]

Hey there, I built something.

I generate a lot of code, docs and analyses with Claude. I got frustrated with the feedback loop that made it hard to give the agent meaningful feedback it could learn from.

You copy-paste comments, you fire up a Google Doc, or you just don't review and ship the slop.

Instead, I built grade (.) dev. A collaborative review tool for AI markdown artifacts. Your agent pushes a markdown doc (prose, code, mermaid) over MCP. You leave comments inline to a specific code line, mermaid node or sentence. The agent reads them back through the same MCP.

I use it to teach Claude my coding style, get a friend input on something, … Kind of a personal RLHF!

If you have been around long enough to remember pastebin: think of grade as the grown up version, built for the agent era.

Free, every review auto-expires. Have fun!

Link👇

[Summation @_summation]

Our customer, @OneLineage runs 250+ warehouses across the global cold storage network.

KPI anomalies were caught reactively.
Investigations took weeks.
Now they take 15 minutes.

👉New case study: https://t.co/G5ZiVPJ0RK

[RC @rcdexta]

# agent-browser skill and memory leaks

These "Google Chrome Helper" processes were hogging my mac cpu for the past few days.

The following did not help:
1. Suspending all tabs on Google Chrome
2. Closing Google Chrome
3. Closing other Chromium based browsers

Turned out there were orphaned agent-browser automation sessions created by Codex / Claude Code with temp Chrome profiles running for days.

Killed the parent processes and my cpu usage dropped to <10%

[RC @rcdexta]

So far the best product execution of bring your laptop agent sessions to the phone!

[OpenAI @OpenAI]

You've been asking for this one...

Now in preview: Codex in the ChatGPT mobile app.

Start new work, review outputs, steer execution, and approve next steps, all from the ChatGPT mobile app. Codex will keep running on your laptop, Mac mini, or devbox.

[RC @rcdexta]

Weekend home networking project: Done✅

5 Gig fiber, new router, switch, cabling, and a few rounds of “why is this link negotiating at the wrong speed?” later… we have liftoff!!

[RC @rcdexta]

My biggest issue with AI-generated analysis is not the prose. The prose is usually great.

The problem is everything underneath it.

Do the numbers tie? Are the claims verified? Can the logic be rerun next month?

That is the bar we built Reports for at Summation.

https://t.co/5zDTeXzlgn

[RC @rcdexta]

Codex already can switch to gpt-5.5, even if not available in the /model selector you can try

codex -c model_reasoning_effort=xhigh -m gpt-5.5

[RC @rcdexta]

This is worth paying attention to. If you’re building agent sandboxes and need save/restore for filesystem state, Amazon S3 Files is a very interesting new primitive. It gives S3 buckets file system semantics without moving data out of S3, which makes shared agent state and persistent sandbox storage much more practical.

https://t.co/CyuLwqHdzD

[RC @rcdexta]

The "Margin Notes" function uses @steipete awesome summarize library to find free models available at anytime: https://t.co/QmyS3gI5Zy

[RC @rcdexta]

New blog post: Prompt Caching from First Principles

Most explanations stop at "identical prefixes get cheaper." This one walks through the full transformer pipeline: tokenization, embeddings, KV caching to show exactly why caching works, what breaks it, and the math that makes the savings obvious.

The twist: the blog has an AI co-author built in, I call it "Margin Notes".

Select any paragraph you find dense, and Margin Notes will reframe it for you, as a more intuitive explanation, an analogy, a step-by-step breakdown, or even a quiz. Same content, shaped to how you think.

https://t.co/8fFf7cGpew

[Matt Harrison @__mharrison__]

For my friends who are still using UV and might be a little weary about recent compromises to PyPi packages, stick this in your pyproject.toml.

You can let all of those pip users find and report the compromises...

[RC @rcdexta]

OpenClaw - Perplexity Edition

[Perplexity @perplexity_ai]

Announcing Personal Computer.

Personal Computer is an always on, local merge with Perplexity Computer that works for you 24/7.

It's personal, secure, and works across your files, apps, and sessions through a continuously running Mac mini.

[Jason Bosco @jasonbosco]

"We used to debate using tabs vs spaces in code we'd type out"

[CG @cgtwts]

Claude Code watching me write code manually

[Thariq @trq212]

You can now connect to Claude in Chrome using the VS Code extension.

Use it to debug frontend apps, collect data or automate your browser.

Install the extension and type @ browser to get started.

[RC @rcdexta]

Codex skill for Claude because why not!

[antirez @antirez]

If you explain how to do it, Claude Code is able to use Codex when it is incapable of solving certain issues. This way you have the best of both worlds.

Here is the skill file that lets Claude use Codex.
https://t.co/qdit9JQSbm