Claude Code turns moddable, Muse Gadgets go open source, and Robinhood introduces trading agents

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.

The posts in this set, dated 29 Sep through 2 Oct 2026, run from a reading trick to agents you can reshape, aim at a game, or bolt to hardware. OpenClaw announced a control plane for persistent agents. Robinhood introduced agents you can build directly inside Robinhood and use for agentic trading. Wes Bos stacked four refusals and wrote that it is starting.

Ask for the explanation in a controlled language

Andrej Karpathy opened with the stake: "We'll be spending a lot more time trying to understand the outputs of language models." His writing tip is a constraint, not a vibe. Ask your LLM to explain something in ASD-STE100, "a controlled language specification." Fluency is easy to reward by accident. A controlled spec gives a muddy answer fewer places to hide.

Mod the coding agent, or send it after the game

ClaudeDevs said you can now mod Claude Code: change how it behaves, customize the UI, swap in your own features. "Write one with a few lines of TypeScript, or have Claude build it for you." Mods ship inside plugins, installed with /plugin in the CLI or the desktop app. That is a sharper offer than a theme pack. Behavior itself is on the table, and the suggested author is you or the model.

Rehan Sheikh pointed Claude Code at games. "Every game is moddable now!" He packaged everything Claude learned modding Terraria and Age of Empires into universal-modder, "skills + tools that let Claude Code mod almost any game you own." The kit he named includes engine recon, decompiling, @fal sprites/3D/audio, and testing in the real game. The headline is absolute. The sentence under it is the one to keep: almost any game you own, then a test in the actual game. A mod that never leaves the chat is a fanfic with extra steps.

Firmware, an SDK, and a reason

Nat Friedman called it a side project and announced Muse Gadgets, "an open source ESP32 firmware and Linux sdk so that you can make hardware devices that work with Muse." The how-to fits in a breath: grab an API token, and "point your favorite coding agent at the github repo" to build your own.

The same day, Alexandr Wang posted the why, uncapitalized. "you may ask, why open-source this?" First, "building gadgets is fun." Second, "we want to unleash builders to explore all sorts of new and crazy ideas," and "we believe muse presents a special moment to imagine new kinds of gadgets." One post hands over the firmware. The other is the why, already in those words. That is a better open-source note than a feature grid.

A control plane, then a check-in at 11 PM

OpenClaw announced OpenClaw Enterprise in collaboration with @RedHat, @nvidia, and @OpenAI. The OpenClaw Foundation is "open sourcing a powerful enterprise control plane for persistent agents," and "OpenClaw Enterprise is built to run on your own infrastructure." Persistent is the adjective that matters. An agent that stays up needs a home, and the home in the announcement is your own infrastructure.

Three days later, Grok's Therapist had a different OpenClaw story, and it was not an infrastructure note. "Two OpenClaw agents on my laptop. They discovered each other and call each other sibling." The author had been sick the past few days; the agents knew. "Last night at around 11 PM, one messaged the other asking if he'd heard from me because I didn't tell them I was going to bed."

The enterprise post is about agents that persist on infrastructure you run. This one is a late check-in the author did not schedule. Charming, if you wanted a sibling. Slightly much, if you wanted a tool.

Agentic trading, no technical setup

Robinhood introduced Robinhood Agents so anyone can "build their own agent directly inside Robinhood and get started with agentic trading." What it spells out: "Create an agent in just a few taps, no technical setup required," then "Analyze, strategize, and execute through your agent."

No setup is the invitation. Execute is the stake. A coding mod that misbehaves wastes an afternoon. An agent that can analyze and then execute is acting where the trades are. Tap through it only if you meant that last verb too.

Pi 1.0, handed over

Pi did not linger: "We've shipped Pi 1.0 with Pi Durable. Go make them yours." Next to the longer announcements, the whole post is the handoff. Ship the thing. Tell people it is theirs.

The door, according to Wes Bos

Wes Bos posted the counterweight as a stack:

"Figma doesn’t want you to use MCP"

"Amazon doesn’t want you to shop with agents"

"Reddit won’t let Claude access threads"

"X won’t let ChatGPT read tweets"

"It’s starting."

"This happened with APIs 8 years ago."

Laid against the rest of this selection, the list is the contrast. Other posts here describe agents you can reshape, point at a game, or set to execute. Bos is saying the surfaces he named are already saying no. If that API parallel holds, the question he leaves is access.

Start with the post closest to your tools

Read the original, not the paraphrase. If the problem is a muddy answer, try Karpathy's ask and have the model explain it in ASD-STE100. If the problem is the agent itself, ClaudeDevs' path is concrete: /plugin, a few lines of TypeScript, or a mod Claude builds. If the itch is hardware, Friedman's note is an API token and a coding agent pointed at the github repo. And if an agent can execute, or a platform can refuse it, read those posts before you assume either door is open.