Some @bot tags on X never reach your Bot; staff are tracking it

Grok BotWhat's new#

No new Grok Bot version is on the changelog; it still ends at 0.70.2 from Oct 11. Cursor's status page is All Systems Operational this morning. (confirmed)

Some @bot tags in X replies never reach your Bot (confirmed)#

What happened: At 8:47 PM ET on Oct 14, Cursor staff member kevinn said two @bot reply tags a user sent never reached their Grok Bot: "they were stopped before the handoff rather than lost inside your Primary Bot." He called it "a known issue with tags on replies that we're already tracking." A tag is skipped "without any reply from @bot" when the post or a quoted post has a video or GIF, when @bot is only in the reply names X fills in, when @grok is tagged too, or when weekly usage is used up.

Who is affected: anyone who sends work to Grok Bot by tagging @bot on X, most of all in reply threads where @bot was already tagged above.

What to do now:

1. Look for an @bot reply such as "On it!" under your post. No reply means the tag never reached your Bot.
2. Type @bot in the text of your reply, not just in the reply names.
3. Leave @grok out, and skip posts with a video or GIF.
4. If @bot still doesn't reply, paste the post link into your Primary Bot's chat with your instruction.

What to watch: the reporter later said @bot was typed in the body of both replies too, so step 2 may not always be enough. Staff haven't answered that yet.

New grok_ API keys run CLI and SDK agents again (confirmed)#

What happened: At 5:36 AM ET, Colin said "the change that fixes agent runs with the new permission-based API keys (the ones starting with grok_) is now live." Since Oct 13, those keys could sign in and list models but every agent run failed with Service Unavailable.

Who is affected: Cursor CLI and Python SDK users who created a new permission-based key; the original reporter had just moved to The X Subscription.

What to do now: set CURSOR_API_KEY back to the new key and run a one-line cursor-agent -p check. Colin says it "should return KEY_OK," and "you no longer need the older key or the browser login workaround."

What to watch: any reports in the thread of keys still failing.

Staff fill in yesterday's computer outage (confirmed)#

Colin said at 5:35 AM ET that "From about 13:00 to 21:00 UTC on October 14, many Bot computers stopped responding," about 9 AM to 5 PM ET. During that time "Update can't make the backup it needs before moving your computer, so it shows "Your computer's files aren't ready" and leaves everything as it is. That was resolved last night." If a Bot still can't use its computer, quit and reopen Grok Bot first, and keep Recover and Reset last.

Grok BotHow people are using it#

• A skeptic's first week. NetworkChuck, a tech YouTuber, posted at 11:48 AM ET Oct 14: "I was wrong about @bot. I said I didn't care. A week in, I get it now. I connected my email, calendar and custom apps, and it just works." X shows about 34 million views, and @poteto reposted it at 10:01 PM ET. (reported, self-described)
• Booking a flight, with a safety note. @XFreeze said at 1:39 AM ET their Bot booked a flight and "I send a photo of my credit card and it did the entire process." @poteto reposted it. Our note: a card photo stays in your chat history. Since v0.68.1, "checkout card fields get a Secure Form instead of a screen handoff," so let the Bot ask for the card in that form instead. (reported, one user)
• A Bot that watches for phishing. Paul Vann (@pjvann), who works at SpaceXSI per his X bio, shared a template called "Am I being Phished?" that "watches your email, @chat, and slack for any suspicious messages or requests, and flags them in real-time." @poteto reposted it. (reported; builder is SpaceXSI staff)
• One Cursor Project per workstream. Vincent van der Meulen (@vinvan) described his setup: "create a (cursor) project for every big workstream. i have ten," have each project start cloud agents with /poteto-mode, and keep an engineering Bot in Grok Bot. @poteto reposted it. (reported, one user)

Beyond Grok Bot#

• Anthropic: Claude Platform can now send OpenTelemetry traces of Messages API calls, Managed Agents sessions and the Claude SDK to your own observability tool, in beta. [[entry:claude-platform-opentelemetry-traces]] (confirmed)
• OpenAI: eight states sued OpenAI on Oct 14 over ChatGPT's safety for kids; here is what ChatGPT's parental controls can and can't do today. [[entry:eight-states-sue-openai-chatgpt-safety]] (confirmed)
• Hark: Brett Adcock says Handoff passed 1 million browser sessions for Hark Pro users, an unaudited figure, plus how people use it and where it stumbles. [[entry:hark-handoff-one-million-sessions]] (reported)

Thursday showcase#

Five weekly AI builds you can copy, from client recaps to Canvas grading, each cited with a starter prompt. [[entry:showcase-best-builds-oct-15]] (reported)

---

Grok BotThe digest#

• Idle Bot desktops will close on their own, but not yet. Colin said at 5:50 AM ET that the team "has finished a change that closes a bot's desktop after it has been idle for a while, and also keeps a desktop's old processes from being left behind when it restarts." He added: "It isn't switched on yet, so I don't have a date to share." Until then, he says spreading out routines that wake many Bots at once and asking Bots to close their browsers when they finish still help. (confirmed, staff)
• Cloud-only notes for Cursor cloud agents go in .cursor/CLOUD.md. Colin said at 6:08 AM ET that cloud agents read .cursor/CLOUD.md and "It is only loaded for cloud agents, so agents in the IDE never see it." His suggested split: keep the app map in AGENTS.md, put cloud setup and test notes in .cursor/CLOUD.md, and keep both short because "Whatever is in them gets re-read at the start of every run, so a lean file is also cheaper." (confirmed, staff)
• Cursor CLI over ACP can silently use Grok 4.6 Fast. deanrie said at 6:29 AM ET it is "a known issue we're tracking": when an ACP client connects without Cursor's advanced model picker, choosing the non-Fast model is ignored and "the requests are actually being sent to the Fast variant." The workaround is to send "_meta": {"parameterizedModelPicker": true} in clientCapabilities, then set the fast config option to false before the first prompt. (confirmed, staff)
• Cursor CLI over ACP reads subagents, rules and skills from the folder it was started in. deanrie confirmed at 3:28 AM ET that the CLI looks in "the folder where the agent acp process was started, not in the project directory the client passes in session/new," and passed that to the team. Start agent acp from the project folder for now. He also said the CLI "does not load user-level subagents yet," so keep copies in the project's .cursor/agents folder. (confirmed, staff)
• Origin Projects that were renamed can miss their environment secrets. Colin said at 6:43 AM ET that one user's Project kept starting agents on the temporary repo name it had before it was named, so "secret injection is skipped for them." He flagged it and asked the user to test a new Cloud Agent started directly on the repo. He also noted secrets save from the Secrets section, while the Save button only covers install and start scripts. (confirmed, staff; one account)
• You can't limit a connector to one Bot yet. deanrie said at 6:32 AM ET that "per-bot access to connectors isn't available" and connectors apply to every Bot on the account. Bot-level scoping "is something the team has on the radar," with no timeline. For now, add a second labeled account to a connector and tell each Bot which one to use, though "this doesn't strictly limit which accounts the bot can see yet." (confirmed, staff; wishlist candidate)
• Basecamp says its official Grok Bot connector is live. Jason Fried, who runs 37signals, posted at 4:36 PM ET Oct 14: "The official Grok @bot connector for Basecamp has launched!" He says to open Connect Apps and search for Basecamp, and that the listing may show the 37signals logo for now. @poteto reposted it at 8:48 PM ET. We have not seen it on a Grok Bot page yet. (reported, partner claim)
• Stale phone registrations can't be cleared by hand. In the thread on missing Grok app notifications, Colin replied at 5:31 AM ET: "We cannot clear these manually." There's no fix date in the thread. (confirmed, staff; short reply)
• macOS updater follow-up. On the long Cursor update thread, Colin told a user at 5:49 AM ET that a "Domain not found" message from the cleanup step "is harmless" and to check Cursor > About Cursor for version 3.24.18 or later. (confirmed, staff)
• Grok app can't load your Bots after switching to The X Subscription. Colin said at 5:20 AM ET that for one user "the Grok app and grok.com are still connected to an older Cursor account that was linked to your Grok login before you switched to the X Subscription. That older account doesn't have a plan, so the Bots page fails to load and you don't get the + button." He called it "a known issue we're already tracking" and said "Your Premium plan and your bots are fine." Until it's fixed, "the Grok Bot desktop app is connected to the right account, so you can create and use your bots there." (confirmed, staff)

Anthropic#

Claude will answer US voting questions with official Democracy Works data. Anthropic's Oct 14 midterms post says that later this month Claude.ai will start drawing official voting information from Democracy Works into its answers for US users, so "where do I vote" gets a state-specific reply, and it will show up-to-date official results on election night and in the days after. Anthropic says "about 0.51% of all conversations are specifically about elections or voting," based on a random sample of 250,000 US Claude.ai consumer conversations from Sept 21 to 27. Anthropic also says it added classifiers for multi-step misuse and banned accounts behind a foreign-run fake news site. Anthropic

Claude Code 2.1.300: OpenTelemetry attributes, cache-write split, and a stuck-MCP fix. Claude Code 2.1.300 landed on npm at 10:27 AM ET on Oct 14. It adds OpenTelemetry GenAI semantic-convention attributes to trace spans and splits cache writes in the api_request event into cache_creation_5m_tokens and cache_creation_1h_tokens. It also adds /advisor on Claude Platform on AWS, where no advisor runs until you pick one. Among the fixes: a conversation no longer fails "on every turn with a 400 error" when the API rejects an MCP tool's input schema. Update with npm i -g @anthropic-ai/claude-code@2.1.300. CHANGELOG

Managed Agents scheduled deployments reach Claude Platform on AWS. The Oct 14 platform release notes add one line beyond OpenTelemetry: "Claude Managed Agents scheduled deployments are now available on Claude Platform on AWS." Scheduled deployments, launched on the Claude API on June 9, let you "run sessions on a cron schedule without managing your own scheduler." Teams that buy Claude through AWS billing can now run those timed agents there too. Release notes

OpenAI#

Codex usage limits reset again on Day 10. Codex lead Tibo (@thsottiaux) posted at 12:55 AM ET: "The reports of my reset button being lost are greatly exaggerated. Day 10/ Reset." A minute later he said it was "Landing within the next 2 hours," and at 2:28 AM ET: "Done and dusted." It's the second reset in three days after Day 8's on Oct 12 to 13. A reset refills your current limits; nothing in these posts raises them. If you also hold a banked reset, OpenAI's June 11 notes say banked resets are "usable for 30 days after they are granted," so check its date before you spend it. (confirmed, staff posts) Day 10 · landing · done · banked resets

GPT-5.5 is gone from Codex with ChatGPT sign-in; API keys still work. OpenAI's changelog confirms the retirement went through "at noon PDT (19:00 UTC)" on Oct 14 for Codex users signed in with ChatGPT. "The OpenAI API and Codex sessions authenticated with an API key aren't affected." The suggested replacement is GPT-6.1 Sol (gpt-6.1-sol). If a scheduled task, custom agent or saved config still names gpt-5.5, switch it now or sign that workflow in with an API key. (confirmed) Codex changelog

OpenAI: agentic tools now make over half of output tokens in its EU business sample. An Oct 14 OpenAI Economic Research report says that in its EU sample of ChatGPT Enterprise and Business accounts, "Codex and ChatGPT Work now account for more than half of output tokens." Weekly active Codex and Work users at EU small and medium businesses (200 or fewer users) "grew more than eightfold" since May. The agentic share ranged "from 22% in Slovenia to 85% in Lithuania." EU-based firms were "only 10% of the frontier firms globally." These are OpenAI's own usage figures, not an independent survey. (confirmed, OpenAI data) OpenAI · report PDF

OpenAI's infrastructure and compute lead Sachin Katti steps down; Uday Ruddarraju takes over. Katti said on LinkedIn he "decided to move on to build something new" and will advise OpenAI after his last day on Nov. 15, CNBC reports. An OpenAI spokesperson told CNBC that Uday Ruddarraju, its CTO of compute and formerly head of infrastructure engineering at xAI, will lead the overall compute organization. CNBC says OpenAI is targeting "roughly $600 billion in total compute spend by 2030." (reported by CNBC and Bloomberg, citing Katti's LinkedIn post and an OpenAI spokesperson) CNBC · Bloomberg Law

OpenAI accuses an ex-employee who returned to Apple of taking device files. In an Oct 14 filing in Apple's trade secret suit against it, OpenAI accuses a former employee, who came from Apple and has since returned there, of sending "over a hundred highly confidential files about at least one unreleased OpenAI device" to his personal accounts. OpenAI also says it understands he made a presentation about its product plans to Apple's senior leadership. It asked for time to investigate and possibly file counterclaims. Apple told Reuters it has found no evidence to support the allegations, and its lawyers said they had found no OpenAI files on Apple's servers so far. Bloomberg Law reports OpenAI agreed at Wednesday's hearing not to use the trade secrets Apple says were taken while the case plays out. No court has ruled on either side's claims. (filing confirmed; claims are OpenAI's allegations; Apple response and hearing reported) Filing · Reuters · Bloomberg Law

Known issue: Costs data delayed in the Compliance API (admins only). OpenAI's status page shows an Identified incident, "Delayed Costs data in the Compliance API," ongoing about 18 hours at our 7:08 AM ET check. "A fix is in progress and currently undergoing testing. API and ChatGPT usage are not affected, and other reporting channels remain available." If you pull workspace costs by API, use another reporting channel or hold today's cost reports. (confirmed) OpenAI status