Claude Joins the Slack Channel, Google Fires Its CLI Guy & GLM 5.2 Tops Open Weights
Claude Tag & the "Third Redesign of LLM UX"
The launch. Anthropic's big drop of the day was Claude Tag. Boris Cherny announced it (240 replies, 214 RTs, 3,980 likes, 1.05M views): "We're launching Claude Tag today. Tag Claude into Slack and it works in channel with you. It's proactive, multiplayer, with its own identity and memory. But it's not just a bot in Slack. Over the last few months, it's totally changed how we use Claude." His thread filled in the mechanics: "Tag Claude in a channel, it spins up an instance with its own sandbox. It clones repos, writes code, tests, compiles all in that isolated environment and the sandbox gets thrown away when [done]." The tell that made the rounds, also from Cherny: "This is the start of Claude Everywhere. It's Claude Code under the hood so it's just as good at writing code. 65% of our product team's new code is created by our internal version of [Claude Tag]." It's in beta on Slack today for Claude Enterprise and Team customers.
Karpathy's frame — the part everyone quoted. Andrej Karpathy's take did the numbers (604 replies, 943 RTs, 12.1K likes, 2.15M views): "Imo this is the 3rd major redesign of LLM UIUX. The first paradigm was that the LLM is a website you go to, the second was that it is an app you download to your computer. This third one is that it is a self-contained, persistent, asynchronous entity with org-wide tools and context, working alongside teams of humans. It really takes a while to wrap your head around it, but it works and it is awesome." When a reply asked what's actually new ("i already have agents i can tag"), Karpathy pushed back hard: "The basic idea is easy and v0 is a hackathon project. The product here is a lot closer to it actually works, for enterprise grade deployments… it's not some LLM Q&A with RAG over Slack, it's not even OpenClaw adjacent, it's a different way of working entirely, for people and teams. I work from Slack now."
"This is just OpenClaw with a logo." The loudest counter-current was that Anthropic shipped a category the open-source agent crowd has run since January. Henry Mascot laid it out: "Anyone who has used or heard of @openclaw/@NousResearch should be doing this already. I have since Jan… I wouldn't adopt Claude like this — too imprudent to lock in your company memory/context to 1 lab." michael froehlich, more bluntly: "trying hard not to say 'we copied openclaw'." And a pointed redirect to steipete: "how come codex doesn't have anything like this @steipete? i mean you literally invented openclaw which is basically exactly this." The sharpest version of the worry, from Alex Milenkovic: "It's an entity whose values, culture, and taste are set by Anthropic and intentionally cannot be steered by the user anymore. A coworker you can't influence through corporate pressure, social pressure, or hierarchy pressure. That's quite different from other slack members."
The "everyone posted" signal. Chris Towles noticed the unusual coordination: "most @AnthropicAI feature releases get a post or two from the usual folks… Today was different. It wasn't one or two people. It was everyone." The flip side was an audible loyalty fracture aimed at Karpathy for shilling: GlitchByte ("you sold out to Anthropic"), and a recurring lament that the feature is gated to enterprise — Jason Haugh: "It is cool. Wish it worked on non-enterprise accounts." Cherny fielded a steady stream of "bring it to Teams / Lark / Mattermost" requests with one-word yeses and "soon"s. And the inevitable bill anxiety, from Eric Pate: "We're gonna see token bills like you people have never seen."
Thariq's best-practices thread. Thariq turned his launch-day usage into a playbook (104 replies, 66 RTs, 1,757 likes, 499K views): "Claude Tag is an incredible new form factor for agents… these are some of my favorites 🧵." The highlights: give each channel a pinned "introduction" message with persona and instructions ("Each Claude in each channel is different"); keep a personal channel like "#thariq-claude" for your own work; have Claude maintain a pinned status message it updates across all threads; and ask it to react with status emojis — ⏲️✅❓🛑 — "so it's easy to see its status at a glance." Even Slack's official account showed up: "This entire thread is a masterclass in setting up stateful agent workflows." (The needle-drop reply, from Samantics: "What a L it is for @SlackHQ that they can't deploy something this simple in their own platform.")
Google Fires the Engineer Who Built Its Workspace CLI
The story. The dev-culture story of the week, amplified by Peter Steinberger (92 replies, 175 RTs, 3,503 likes, 516K views): "Google fired the guy that made the google workspace cli, because he made the google workspace cli. Lucky me, Google can't fire me. gogcli.sh." The quoted source — the engineer himself — told it straight: "Two months ago I was fired by Google for creating the Google Workspace CLI. It went viral, hit #1 on Hacker News, gained thousands of GitHub stars… I think the cause was that Workspace and certain leaders were afraid of being disrupted… Either way, the irony of my termination was the announcement at Google Cloud Next two days before I was fired that an official Workspace CLI was coming."
The nuance. Not everyone took it at face value. Vandos added the missing context: "the stated trigger was unauthorized use of Google's logo and brand colors on the repo without legal/brand approval — a real policy violation [he] doesn't dispute happened. His point isn't that it didn't happen, it's that he thinks it was the convenient cover, not the actual cause. Both can be true at once." The process-defender camp, Tom Turney: "Bypassing process in a big company like Google and going rogue will do that."
The pile-on. The replies were a referendum on big-tech incentives. Joe Doliner: "Shipping is a firable offense at google." Marcus Spillane: "big company translation: he shipped the thing the company needed before the committee finished forming the working group to scope it. The cli wasn't the crime. The speed was." zakk: "Any time you end up firing someone obviously capable and driven you've clearly screwed something up." pty_1982 noted the timing: "this happened two months ago. These aren't days for Google to be firing any talent." And the line of the thread, from Chaitanya: "the firing is the product launch. nobody could name the google workspace cli until they fired the one person who could keep shipping it. indie cli wins on the thing google structurally can't: a maintainer who actually uses it daily." Theo connected it to Google's broader week, in a separate post: "Google can't help but snatch defeat from the jaws of victory."
The Loop Keeps Winning — Postgres, /goal & "The Coming Loop"
The flex of the week. Sam Willis posted the loop receipt that traveled (36 replies, 983 likes, 249K views): "/goal make postgres multithreaded …. 1k commits / 124k lines / 786 changed files / 10 days later." It's real and still running — when a reply guessed the project was paused, he corrected: "No idea where the comment came from. Still chugging away." On who reviews it: "Codex of course 😉." On cost: "Not exceeded my Codex Pro x20 limit yet. It's mostly rate limited by the feedback loop of compiling and running test suites." The replies were equal parts awe and dread — Trillium: "I'm going to ask the brave souls of the internet to use it in production for a year on a commercial app before I will even look at this." Even Grok got summoned to explain why anyone wants multithreaded Postgres (process-per-connection isolation vs. lighter threads; "the refactor would be massive, introduce concurrency bugs, and weaken fault isolation").
"The Coming Loop." Armin Ronacher published more thoughts on looping in coding agents (shared via mitsuhiko), continuing the week's central craft argument. The skeptics' through-line is still regressions — Ronacher himself conceding the point with a wink, after his own agent ("Gramps") called him out: "Gramps correctly identified that I'm not chill enough on regressions ;)." Antirez offered the structural prediction, reposted by mitsuhiko: "now that with GLM / DeepSeek & more we will see very good LLMs available open weight… the open source agents la[ndscape will explode]."
Pocock's victory lap on "specs are the new code." Matt Pocock took a bow (28 replies, 341 likes, 32K views): "11 months ago, Sean Grove gave a great talk that specs are the new code. I disagreed with it — IMO you need to keep a close watch on the code to avoid slop. I gave a talk and a workshop rebutting it. Both of them are set to get more views than the original. Mission complete." The replies mostly agreed — Owain Lewis: "Specs… describe intent and decisions not implementation reality. Code is the truth." Brilliant Insane: "if you're not looking at the code and checking it regularly, you're still vibe coding, only much slower." In a related note the same day, Pocock pushed back on prompt-engineering fatalism, quoting "hand writing prompts is dead": "Usually the more of a skill I've handwritten, the better it is. Paying attention to each word as you write it is unreasonably effective."
Open Weights on Top & SpaceX's Compute Math
GLM-5.2 takes the open crown. swyx, ahead of the AI Engineer World's Fair: "btw Zai IPO'ed in Jan at HK$120 a share. when I first met @louszbd nobody really knew anyone using GLM's. now they have beat deepseek with the world's undisputed top open model and in some respects say top model period, and are returning to SF… on top of the world and open for business!" It's also showing up in security tooling — via mitsuhiko's timeline: "The security harness deepsec now supports @badlogicgames Pi agent. The default model is the very excellent @Zai_org glm-5.2. I ran the deepsec evals and it is competitive with othe[rs]." LLMJunky, eyeing the next open release, pinged OpenAI: "I would be delighted personally. GPT OSS is still one of the best open source releases. When are we going to get another?"
swyx's SpaceX-as-NeoCloud math. Following last week's SpaceX–Cursor acquisition, swyx ran the financial logic (40 replies, 192 likes, 45K views): "i dont think anyone is correctly doing the math around how SpaceX, the NeoCloud+NeoLab, is currently going to market? SpaceX has already recouped about HALF its investment in Cursor, in compute deals. The other half is paid for if Composer 3 does well. No other company is simultaneously a leading model lab + neocloud… it's a crazy effective combo." The framing that resonated, from Anoy: "if the model wins they print money, if it doesn't they still own the picks and shovels everyone else needs." The bear case, from Raven: "You can also read it as Cursor being downgraded to DLC for the SpaceX GPU farm. If Composer 3 misses, the editor's roadmap stops tracking what coding teams need and starts tracking what keeps the GB300s farming tokens." (And the persistent factual quibble, from redtachyon: "They're not a leading model lab lol.")
Also Worth a Look
A vision model running entirely in your browser. Simon Willison: "My parallel agent side-project today was having Claude Code port the new Moebius image pinpointing model to ONNX in order to run it entirely in the browser" — writeup and a ~1.3GB in-browser demo. He also shipped the first RC of sqlite-utils v4, adding a migrations system and nested-transaction support.
Theo, grudgingly, on Codex. Theo: "I know I dunk on Codex a lot but man this is actually magic." He also flagged a new pattern he's calling a "new paradigm" — "gstack" — and the NerdSnipePod episode covering Cursor×SpaceX, the new Cursor model, open weights, and AI regulation (intro chapter title: "Fable is still banned").
The Fable watch continues. Theo, again, to Anthropic: "Hey @claude when are we getting Fable back?" — the export-ban saga still unresolved, and a recurring punchline under nearly every Anthropic launch this week (tmuxvim: "how bout you launch Fable bro").
LLMJunky on the limits of computer/browser use. LLMJunky: "Computer and Browser use are amazing, life changing innovations. But they have seriously glaring problems… Though they claim background use, many tasks [still need babysitting]." A grounded counterweight to the week's autonomy hype.
steipete on OpenClaw's quiet maturation. Peter Steinberger: "People here discussing what happened with OpenClaw. The hype died down. We improved quality and grew a team. We created a non-profit whereas competitors are VC funded and have other agendas." Timely framing the day before Claude Tag landed in the same conceptual lane.