Cloud sessions leave preview, Copilot leads with Autopilot, and Muse teams with Shopify

X Post Facto is our weekly pause on the feed: a look back at the AI posts that stood out, what they said, and why they’re worth a closer read.

This selection is a product week with a workshop attached. Cloud sessions for Claude Code left research preview. Satya Nadella called a Copilot update the biggest yet and put a long-running enterprise agent at the front of it. Muse showed up three ways: built from scratch and inspired by openclaw, paired with Shopify on shopping and checkout, and flagged as a place you can install Claude Code. Around those posts sat the notes you would actually paste into a session — a cache caveat, a pre-ship pass, a skill for talk materials — plus a Grok 4.7 demo and a sqlite regret.

Cloud sessions, lid optional

On 23 September, ClaudeDevs said cloud sessions are "officially available and out of research preview". They "let you keep Claude Code working, even when your laptop is closed." Existing subscribers get a one-time credit to try them: $100 on Pro, $250 on Max.

The credit is the more interesting line. A one-time allowance to try the feature is a trial, which is how you learn whether a session that survives a closed lid is something you actually leave running.

Copilot, as an OS, led by Autopilot

Satya Nadella wrote on 25 September that they are "building Copilot as a new OS for work that spans every model, every form factor, and every task." He called the day's news "our biggest update to Copilot to date, bringing four things together", led by Autopilot, a "proactive and long-running agent built for the enterprise", and then Code.

"Every model, every form factor, and every task" is the wide shot. Autopilot is the part you can describe without the slogan: proactive, long-running, built for the enterprise. Set next to the cloud-sessions post, the overlap is an agent that continues after you step away, and the difference is the room. One is a closed laptop. The other is the enterprise. They are not the same offer.

Muse, from Mac minis to a checkout

Nat Friedman drew the lineage in one sentence: "we built muse from scratch, but it is definitely heavily inspired as a product by openclaw." After using openclaw in January, he bought "hundreds of mac minis for the MSL team", and "lots of us fell in love with using openclaw (and other personal agents)." In that post he calls @steipete a genius.

"From scratch, but inspired" is worth keeping intact. It refuses a reskin story. The clauses after it still do the other job: he used openclaw, bought the minis, and a lot of the team fell in love with it. Hundreds of Mac minis is a receipt.

The same day, Mark Zuckerberg posted the commercial counterpart: "Teaming up with Shopify to make shopping and checkout easier in Muse. Shoppers find more. Shops sell more. More partnerships like this coming soon."

That middle pair is a pitch, not a tally. The announcement is the partnership; "shops sell more" is what it is supposed to do.

On 26 September, Federico Viticci added the line practitioners will actually test: "I don't think people realize you can install Claude Code inside Muse lol".

That line nests Claude Code inside Muse. Worth trying before you file them as separate tools.

Sync was fine, until the sessions multiplied

Peter Steinberger named "The biggest design mistake I made when we moved OC to sqlite: using sync db access." It was fine "when it was just an agent that reports to you on Slack or iMessage". It is limiting "now that one agent might do 50 sessions in parallel and the whole team works on it".

Friedman had called him a genius on 21 September, in the openclaw post. This note is a different mood. If 50 parallel sessions is anywhere near your setup, borrow the regret before you copy the access pattern.

The Claude Code notes worth pasting

Lydia Hallie, on 22 September: "Switching effort mid-session on Opus 5.5 doesn't break your prompt cache btw! Just make sure you're on Claude Code v2.1.280+"

The version is the advice. On Claude Code v2.1.280 or newer, changing effort mid-session on Opus 5.5 leaves the prompt cache intact. She is not making that promise below the pin.

Vox offered a pre-ship pass on 26 September: before you ship a vibe-coded site, have Opus 5.5 send out 5 subagents to check 20 things, then fix them all in one pass. The first item is consistent design. "Write a DESIGN.md first, and make every color, font size, spacing and corner radius follow" the reference linked in the post.

Five subagents and twenty checks is a lot of faith in one fix-up pass. The piece worth keeping even if you skip the subagent pass is the DESIGN.md: settle color, font size, spacing, and corner radius before anything starts sanding the site into agreement.

みのるん opened the week on 20 September by publishing a skill:

"私が普段、登壇資料をClaude Codeに作らせているノウハウをスキルとして公開しました！

たぶん有料級ですが、お金は興味ないので本買ってくださいｗ"

He is releasing the know-how he usually uses to have Claude Code make his talk materials. He calls it probably worth charging for, then waves the money off and tells you to buy the book. The ｗ does the tone. Same tool as the cloud-sessions post, pointed at materials for a talk rather than a session that outlasts the laptop.

A model trained for the bot

Alex Finn said Grok 4.7 "just released" and is "an EXCELLENT model". "It was trained FOR Grok Bot". He takes that to mean "a fully agentic model trained to do your knowledge work better than you can", and points to a video of a Grok Bot workflow "that will 10x your productivity".

The 10x, and the claim that it will outwork you, are the pitch for the video. The line that sits with the rest of this selection is the training target: a model aimed at Grok Bot.

If you open only a few, start with ClaudeDevs for the credit and the closed-laptop behavior, then Lydia Hallie before you change effort on Opus 5.5. Read Peter Steinberger if synchronous sqlite still sits under anything that might fan out toward 50 sessions. Federico Viticci is the one to try with Muse already open. Alex Finn is the post with the video.