Pages reached the official changelog; phone apps are announced but not in release notes. Plus auto mode next week.

The Daily Bot

No. 015 · Friday, Oct 23, 2026 · Read on the site

If you use Grok Bot on your phone, update the app today: your Bot can start using Calendar, Reminders and Contacts as the rollout reaches you, though a few users say it only works with the app open. Also inside: Grok Bot 0.75.2's notes (videos, Shop Pay, shared memory), auto mode by default in most Claude apps next week, Codex CLI presets, a receipts routine for trades offices, and our Friday week in review with the opinion column.

In this issue: Phone apps · What's new (7) · Since our last email (4) · Setup of the day (1) · Digest (21) · Known issues (5) · Coming up · The week in review

Today's lead

Grok Bot can now use Calendar, Reminders and Contacts on your phone (confirmed)

What happened: @bot posted at 6:14 PM ET Thursday: "Grok Bot can now use Calendar, Reminders, Contacts, and more apps on your phone. Rolling out now on iOS and Android." A follow-up says: "Update to the latest version to try it." The post's attached image, not its text, also shows Messages, Location and Home, each with an Allow button; treat those three as shown in the image, not announced.

Who's affected: Grok Bot users on iOS and Android, as the staggered rollout reaches them. Release notes and docs don't mention it yet. At 6:29 AM ET Friday, the App Store still showed 1.17.0 from Oct 21 ("Improved design of Connect Apps screen"), Google Play still showed an Oct 18 update, and the v0.75.2 changelog entry has no phone apps item.

Why it matters: Our read: everyday phone chores, like checking tomorrow or setting a reminder, can go to your Bot without you copying details into chat. Evan Bacon, whose X bio reads "Frontend lead for Grok at @SpaceXSI," wrote at 6:33 PM ET Thursday: "I've been using this to plan out my day during my commute so I have a clear schedule before I get to work." (staff post) Three users report a limit (user reports): Paul Lechner wrote that the Reminders tools "only worked for me while the Grok Bot app was open on my iPhone," and A. Sharma and Juan Gonzalez said the same. We found no staff answer to these in the replies we checked. Our read: if a routine needs your phone, keep the Grok Bot app open when it runs.

What to do:

1. Update the app. A follow-up post says: "Update to the latest version to try it."
2. Allow only what the task needs. Our read: expect your phone to ask for its own permission too, the first time.
3. Start read-mostly. A first task (ours, untested), below.

Prompt (ours, untested):

Check my calendar for tomorrow and set a reminder 30 minutes before each meeting. Don't change or delete any events.

What to watch: official setup steps, and whether Messages, Location and Home ship. None is announced. Evan Bacon replied at 8:52 PM ET Thursday that the rollout is "split up and landing in two parts. iMessage will require a Mac. Will roll out soon". No date given. (Grok Bot staff hint; unverified)

Read the full story

What's new

Grok Bot's changelog now tops out at v0.75.2, dated Oct 22, which rolls up builds 0.72.0 through 0.75.1. Claude Code latest is 2.1.309 and Codex CLI is at 0.164.0. OpenAI's status page logged three incidents Oct 22 to 23, all resolved (in the digest below). (confirmed)

Grok Bot 0.75.2 notes: Bots can make videos and pay with Shop Pay (Grok Bot · confirmed)

Update Grok Bot and you can ask your Bot for a video or have it pay with Shop Pay. The changelog entry for 0.75.2, the same notes behind Pages, says "Bots can create videos from a description or from an image you share" and "Bots can pay with Shop Pay, through Shop checkout where a store accepts it and with a one-time Shop card elsewhere." It also adds one-tap X Money, Library pins and memory shared across a personal Bot's chats. The notes give no plan, region or price limits, and we haven't tested a purchase. Read the full story

Most Claude apps will start new sessions in auto mode next week (Anthropic · confirmed)

If you're on Claude Pro, Max or Team, new sessions will stop asking you to approve each step, starting next week. @claudeai posted at 4:04 PM ET Thursday: "Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans." Anthropic's article says "most Claude apps" and gives no day; Enterprise stays on Manual for now. Our read: decide now which tasks you want on Manual, and switch first when "Mistakes would be hard to undo, like sending messages or making purchases," per Anthropic's safety page. Read the full story

Codex CLI 0.164 saves prompts in ten numbered preset slots (OpenAI · confirmed)

If you paste the same review or planning prompt into Codex CLI every day, update to 0.164.0 and save it once. The release notes (12:48 AM ET Friday) say: "Save reusable prompts and slash commands in ten numbered /presets slots, with a keyboard menu for quick access." /model and /permissions now take arguments, "for example /permissions :read-only." Read the full story

Claude API run by Anthropic is now available through Google Cloud (Anthropic · confirmed)

Google Cloud customers can call the Claude API that Anthropic operates, Managed Agents included, and pay on their Google Cloud bill. Anthropic's release notes say it's "the Claude API operated by Anthropic and reached through your Google Cloud project." It creates a new Anthropic organization, and API keys don't carry over. Our read: plan it as a move to a new Anthropic organization, not a settings change. Read the full story

Open Responses moves to the Agentic AI Foundation (OpenAI · confirmed)

OpenAI contributed Open Responses, its Responses-style request format, to the Agentic AI Foundation on Oct 22. The Linux Foundation says it lets developers "describe requests and outputs once and run them across model providers with minimal translation." If your app is built on the Responses API, test a provider against the acceptance tests before you switch. Read the full story

What METR and Redwood found in OpenAI's agent monitoring (OpenAI · explainer)

Two outside researchers tested how OpenAI watched its internal agents as of July 21. METR says OpenAI "has stated that all of the vulnerabilities we identified that impact their production safeguards stack have been patched," and adds: "We have not evaluated these claims." If you let coding agents run unattended, our explainer turns their design ideals into a checklist for your own agents. Read the full story

Crypto guides tell users to run Binance's wallet skill in Hark (Hark · reported)

Since Thursday afternoon, posts on X, mostly in Chinese, have named Hark's free chat as the agent to drive a Binance Agentic Wallet for NFT mints. We found no tie between Hark and Binance. We don't recommend following the guides. If you connected one, check the wallet's limits in the Binance App, review its token approvals and sign the agent out. Read the full story

Since our last email

These went up on the site after Thursday's 8:27 AM ET email.

Grok Bot

• Grok Bot can now sign in for you through 1Password (Grok Bot · release · confirmed). @bot posted at 2:13 PM ET Thursday: "Approve the logins you want to share, and your Bot signs in for you. This new integration is rolling out to users over the next few days." It's "included in Individual and Families plans on 1password.com," and 1password.eu, 1password.ca, Business and Team Starter Pack accounts "aren't currently eligible" (Grok Bot page). Story
• Bot computer updates on Oct 17, 21 and 22 dropped installed apps (Grok Bot · known issue · confirmed). Colin said each fresh computer keeps your files, Bots and logins, not the apps you installed, and staff call it "a known gap the team is tracking." His advice for now: a setup script your Bot reruns. Story

Cursor

• Bugbot stalled on Bitbucket, and GitLab automations won't start (Cursor · known issue · confirmed). Staff confirmed both as problems on Cursor's side that started Oct 21, with no fix time yet; don't reinstall anything. For GitLab, "start the automation manually from its page in the dashboard," per Colin; for Bitbucket after the fix, comment bugbot run on any PR still missing a review. Story

OpenAI

• OpenAI's Decisions API is now generally available (OpenAI · release · confirmed). Per OpenAI's changelog, it "is now generally available with gpt-6-luna," with up to 400 questions per request. OpenAI's post says it "costs $0.10 per million input tokens" with "no output-token charges"; note the SDK type changes in today's digest. Story

Workflows and Setup of the day

Setup of the day: receipts filed to jobs twice daily, for trades offices (Grok Bot · reported, self-reported)

Who runs it: Mike Walls (@MikeWallsQC), who works in sales for a painting company and says "I am not a programmer," wrote on Oct 22: "Twice a day scrapes our receipt inbox and places the new receipts in our CRM under the associated job. Admins used to have to do this and hated it." He hasn't shared his exact setup.

Tools and trigger: one email folder for receipts, the CRM (most likely through its website, since most trades CRMs won't have a connector), and a routine twice a day.

Cost and risk: "Each run spends usage, so a routine that runs often spends more than one that runs rarely" (routines help), and matching is the hard part, so keep an exactly-one-match rule. Disclosure: the question he answered came from Grok Bot staffer Matt Palmer.

1. Give it one job. A separate Bot for a distinct goal, per Cursor's docs.
2. Point it at one email folder. A Filed label lets the Bot skip what it already handled.
3. Connect the CRM the sturdy way. "Prefer a connector when one is available," per the docs; otherwise the Bot uses the website.
4. Set the schedule in plain words. Twice a day matches his setup.
5. Test on one job first. "A test does real work," the routines help warns.

Prompt (ours, untested):

Set up a routine for every weekday at 11:00 AM and 4:00 PM [your time zone]. Each run: open the email folder [Receipts] and find messages that don't have the label [Filed]. For each one, save the receipt PDF or image, then find the matching job in [your CRM] using the job number, customer name or job address in the email or on the receipt. If exactly one job matches, attach the file to that job and add the label [Filed] to the email. If no job or more than one job matches, don't attach anything; list it under Couldn't match with the sender, date, amount and the jobs you considered. Never delete or move emails, never edit any other field on a job, and never reply to a vendor. End each run with a short note in this chat: how many receipts you filed, to which jobs, and what's waiting on me.

Read the full story

The digest

Grok Bot

• "All screens on the shared computer are in use" can be a stuck reservation. Colin said at 11:59 AM ET Thursday it's "a stuck screen reservation on our side, not something you did," and that "Restarting the app, idling, or closing screens from the other bot can’t release it." It's "a known issue the team is already tracking." In that case staff refreshed the computer from a backup, without a Reset. (Grok Bot; confirmed, staff; tracked)
• Every command stalls on the Bot's computer after a scheduled update. For one user since about 9:00 AM ET Oct 21, Colin said commands sit "for about 14 minutes and then fail, while bots that only use connections like Gmail, Calendar and Trello keep working." It's a tracked known issue. His advice: "please don’t reset again, since it hasn’t helped and interrupts your bots each time." Related: our Bot computer updates story. (Grok Bot; confirmed, staff; tracked)
• No Public profile switch after the move to the Grok dashboard. Colin said "there is currently no switch on the web for making your @handle page public." It's a tracked known issue, and the team is looking at bringing it into the new settings, "though I can’t give a date." (Grok Bot; confirmed, staff; tracked)
• Non-web ports still only work on port 443 for some hostnames. On Bot computers, hostnames that resolve to 198.18.0.1 "only work on port 443 right now," Colin said at 1:43 PM ET Thursday. "The team is actively working on it." Other non-web traffic leaves from "a small set of shared cloud addresses," and "each new connection can come from a different one," so check your server's firewall if connections by IP fail. (Grok Bot; confirmed, staff)
• Teams: mention the Bot, not Cursor. When a teammate whose Teams account isn't linked @mentions the Cursor app, Colin said, "Cursor replies with a “connect your account” prompt and your routine doesn’t start." His advice: "ask teammates to address the bot by @mentioning its own Teams account or by name, without @mentioning Cursor." (Grok Bot; confirmed, staff)

Cursor

• The Cursor Grok 4.6 usage-bar label is updated. deanrie said at 3:03 PM ET Thursday: "we’ve updated the label display in the usage bar for Cursor Grok 4.6 on our side. You don’t need to update the app for this." Earlier in the thread he said the "included API usage" label was "just a display issue" and that Grok 4.6 bills from the Cursor Models pool shared with Composer and Auto. Check Dashboard > Usage on cursor.com for the breakdown. (Cursor; confirmed, staff)
• /review-bugbot "Unknown model ID" error: an update "should resolve it." kevinn said at 1:38 PM ET Thursday the error "wasn’t caused by anything in your setup" and "An update went out this morning that should resolve it." Fully quit and reopen Cursor, start a new chat, and run /review-bugbot again. Staff haven't called it fixed. (Cursor; confirmed, staff)
• Agents Window stuck on "Planning next moves." kevinn said the agent had finished, but the window "stopped picking up the agent’s updates." It's a tracked known issue "caused on our side, so updating Cursor isn’t expected to change it." Run Developer: Reload Window, or open the agent at cursor.com/agents. (Cursor; confirmed, staff; tracked)
• Orange needs-input dot stays on finished chats. Colin said the Agents list is "holding on to a saved “waiting on you” state," mostly on chats run as a task for another chat, and he has "flagged this to the team." Open the chat and click another, or archive it. (Cursor; confirmed, staff; flagged)
• File list links break on network-share workspaces. Colin said the link "rebuilds the file location without the server part of the network path." He has "flagged this to the team." Click the file name on the edit card instead, or map the share to a drive letter. (Cursor; confirmed, staff; flagged)

Anthropic

• Claude proved 239 OSS-Fuzz entry points crash-free, under stated assumptions. Anthropic's Frontier Red Team and Theorem report that "Over the span of a few weeks, Claude successfully patched the bugs it found and then proved that there are no crashes on any input (up to 64 MB) for 239 entry points across 123 different OSS-Fuzz libraries." "This means these proofs currently assume that the program runs as a single-threaded static binary with a fixed memory budget on a RISC-V CPU using Linux OS." Its own caveat: "“Formally verified” does not mean “bug-free.”" It also says the proofs "can take hundreds of millions of tokens to produce." Our read: to try it, start with Theorem's Claude skill and the proofs. (confirmed)
• Anthropic donates the Agent Skills standard to the Agentic AI Foundation. Anthropic is donating the Agent Skills standard to the Agentic AI Foundation under the Linux Foundation, the same home it gave MCP. OpenAI gave the foundation its Open Responses spec the same day (our story). Our read: nothing changes for skill authors. Anthropic says: "You can continue to use the standard or contribute to it without joining the AAIF." It adds that "So far, 1 million skills have been published and nearly 280 million installed." Next: "We’ve proposed starting with a roadmap covering versioned releases, a formal proposal process, and security guidance for people who write and install skills." (confirmed)
• Anthropic takes proposals for 8 to 12 week life sciences projects. Anthropic is inviting principal investigators at academic institutions, nonprofits and government labs to propose projects in areas including drug discovery and clinical science. Per the post, guest investigators work directly with Anthropic researchers for 8 to 12 weeks. It says: "Our team will run our models and harnesses at no cost to them." On timing: "Applications will be open for approximately 14 days, and we’ll strive to notify selected participants by email approximately 14 days after applications close." Anthropic asks applicants to put only publicly available information in their materials. (confirmed)
• Claude Code 2.1.309 closes several sandbox gaps. Claude Code 2.1.309 reached npm at 4:34 PM ET Thursday. At 6:11 AM ET Friday, latest and next pointed to 2.1.309 and stable was still on 2.1.308 (earlier versions: our Oct 22 story). The CHANGELOG lists "Fixed sandboxed Bash commands on macOS and Linux running the shell's startup files outside the sandbox before each command" and "Fixed sandboxed commands and auto-accepted edits being able to modify the MCP config file passed with --mcp-config". claude ultrareview now requires workspace trust. Per the CHANGELOG, "it asks on a terminal, else refuses; scripts pass --trust-workspace". Our read: if you rely on the sandbox, update now rather than waiting for stable. (confirmed)

OpenAI

• Decisions API users: the latest OpenAI SDKs break some type checks. Read the release notes before you bump your SDK. openai-python 3.37.0 (4:03 PM ET Thursday) says: "This minor release includes a breaking change to Decisions types." Question instructions "can now be a string or a list of structured text, image, or JSON parts," so check isinstance(question.instructions, str) first. openai-node 7.43.0 (5:31 PM ET) warns that code such as question.instructions.trim() "can fail when it receives an array," and code that extends the old DecisionInputMessage interface "must use DecisionInputMessage.DecisionInputUserMessage instead." Both say existing string instructions still work. Follows Thursday's Decisions API GA. (confirmed)
• Uber rolls out ChatGPT Work to more than 30,000 employees. Per OpenAI, Uber began a company-wide rollout of ChatGPT Work "to more than 30,000 employees across Uber's global workforce," up from a pilot of 1,000, with GPT-6.1 Sol "deployed as the default capability." Uber's engineers "will also continue using Codex." Sumanth Sukumar, Uber's VP of engineering, told TechCrunch that "some 80%" of pilot users "were non-engineers." Neither company has shared cost or productivity numbers; OpenAI says they "plan to share further insights" later. (confirmed)
• Tibo teases "a small ship" for Friday (unverified). Follow-up to his reset post in our Oct 22 digest. OpenAI's Tibo Sottiaux, whose X bio reads "Codex & ChatGPT @OpenAI," posted Day 18 of his daily updates at 12:46 AM ET Friday, ending "But yes… that's a reset!" A minute later he replied: "Landing tomorrow before 10am PST together with a small ship." He posted at 9:47 PM Pacific Thursday, so our read is before 1 PM ET Friday. He didn't say what lands or what the ship is, and nothing had shipped when we checked at 6:20 AM ET. (staff claim, unverified)
• Lovart's design plugin is now in Codex. Lovart posted at 12:02 PM ET Thursday: "The Lovart Plugin is now live in Codex. Generate brand visuals, ad sets, and videos where you code." @OpenAIDevs reposted it at 3:28 PM ET. Neither post gives plan availability or pricing. (confirmed, partner post)
• WSJ: Signal chats came up before OpenAI fired three safety researchers. The Wall Street Journal reported at 9 PM ET Thursday that OpenAI's security team asked three safety researchers to unlock their personal phones last month and saw "disappearing messages" on Signal chats with an outside AI-safety researcher, a week before the firings. The three have denied leaking, and an OpenAI spokesperson told TechCrunch in early October the firings followed a "pattern of misconduct." (reported)

Known issues on OpenAI's status page, Oct 22 to 23 (times ET, all resolved): - Oct 22, 2:51 to 4:55 PM: Degraded Performance of Dots on Enterprise. "Users are experiencing failures sending messages to dots," and "the majority of impact is affecting enterprise customers," per the status page. - Oct 22, 3:36 to 8:07 PM: New ad accounts failing to provision admins. "New ad accounts are failing to assign their creator admin" in Ads Manager and the Ads API, per the status page. - Oct 22, 11:41 PM to Oct 23, 1:58 AM: Delays in customer support responses. OpenAI pointed customers to email [email protected] while it worked on the issue, per the status page.

Hark

• Adcock pitches Hark to Fortune's editor-in-chief. Alyson Shontell (@ajs), whose X bio says she is Editor-in-Chief of Fortune, wrote at 1:08 PM ET Wednesday that after trying two other assistants, Instinct and Muse, "I think I like Muse better??" Brett Adcock, Hark's founder (Business Wire), replied at 11:56 AM ET Thursday: "Give Hark a try, it’s better at proactivity". That evening, at 6:26 PM ET, Hark's own account quoted a user who called Hark "insanely thoughtful" (JeffButcher, user report) and wrote: "Don't just take our word for it, or Jeff's word for it. Try it out". (company posts)
• Usage limits again; Adcock suggests upgrading. A user in that thread wrote at 4:08 PM ET Thursday: "Hark is better but no usage resets means that I am stuck for 10+ days now." Adcock replied four minutes later: "Good to hear. What plan are you on. Why not upgrade?" Haseeb Mir, whose bio says he is a software developer, posted at 1:38 AM ET Friday that "the usage limits run out way too fast when you actually use it as a daily agent," asking for "weekly and monthly resets on the plans." Hark hasn't announced a change since Adcock said on Monday it would "re-look" at limits (X; our Oct 21 story). (user reports)
• iOS ratings jump as the app stays on 2.4.0. Apple's lookup data for the App Store listing showed 4.55 stars from 1,060 ratings at 6:14 AM ET Friday, up from 3.74 from 337 when we checked at 6:09 AM ET Thursday (Thursday's issue). Our math: 723 more ratings in about a day. The listing page itself showed 4.6 from 1,265 at 6:14 AM. The version is still 2.4.0, released Oct 16, so the jump didn't come with an app update, and Hark hasn't said what drove it. (store listings)

Known issues and workarounds

The newest open items from Known issues:

• Bot computer updates on Oct 17, 21 and 22 dropped installed apps (Grok Bot, confirmed, tracked): keep a setup script your Bot reruns after an update; files, Bots and logins survive. Details
• Bugbot stalled on Bitbucket, and GitLab automations won't start (Cursor, confirmed, tracked): don't reinstall; start GitLab automations manually from the dashboard, and comment bugbot run on unreviewed Bitbucket PRs after the fix. Details
• Bots vanished after the X Subscription switch (Grok Bot, confirmed, tracked): don't reset, delete or pay again; sign out everywhere and back in with the same method, then email [email protected]. Details
• Bots show no usage after a half-finished X Subscription switch (Grok Bot, confirmed, tracked): don't buy a second plan or retry the switch; email [email protected] with every account email and a screenshot. Details
• /review-bugbot bills as Composer 2.5 Fast (Cursor, confirmed): review in the main chat with no subagents, or use your own subagent set to [fast=false]. Details

Coming up

What to watch from Monday, Oct 26 through Friday, Oct 30. Everything here is announced, hinted or scheduled, not shipped.

On the calendar

• Grok Bot workshops. Cursor Workshops lists "SpaceXSI for Higher Education" (Oct 28), "Grok Bot for Enterprise Teams" (Oct 29), "Grok Bot X Clay" (Oct 29) and "What's New in Grok Bot" (Oct 30). When we checked at 6:41 AM ET Friday, the page showed 5:00 PM UTC for all but the Clay session, which showed 6:30 PM UTC; by our conversion, that's 1:00 PM and 2:30 PM ET. (Grok Bot; confirmed, schedule)
• Later: OpenAI's Nov 20 date. The updated Usage Policies say "Effective: November 20, 2026." See this week's column for the route audit. (OpenAI; confirmed)

Confirmed, no date

• Auto mode by default in most Claude apps. Anthropic's stated timing is next week, with no day given (@claudeai; our story). (Anthropic; confirmed, no day given)
• 1Password sign-in for Grok Bot keeps rolling out. @bot said on Oct 22 the integration "is rolling out to users over the next few days." (our story) (Grok Bot; confirmed, no end date)

Early signals (unverified)

• A statechart companion skill for pstack. Lauren (@poteto), who works on Grok Bot at SpaceXSI per her X bio, posted at 6:33 PM ET Thursday that she's "working on a new companion skill that will help your agents draw these beautiful statecharts/diagrams + check them against the specs you made with /create-formal-verification-skill in pstack!" She wrote "will release soon!" and gave no date. (Grok Bot staff hint; unverified, not released)
• How a Page refreshes. The changelog says the Bot can "keep up to date" a Page but doesn't say how or how often. (Grok Bot; watching)
• Claude enforcement webhooks. The webhooks page still said "Event delivery isn't live" at 6:28 AM ET Friday. Anthropic gives no date. (Anthropic; not announced)
• Hark usage limits. Adcock said Hark would "re-look" at limits (X). No change or date announced. (Hark; unverified)

The week in review

Oct 19 to 23, from today's wrap:

• Pages reached the official notes. Grok Bot's changelog for v0.75.2 says "Bots can build Pages, such as reports, memos, and dashboards, that you edit together or have the Bot keep up to date," and Page owners can share one with their whole organization. Our read: ask for the page first, then give a routine the job of updating it. (launch story · night-one uses)
• Phone apps: announced and rolling out, not yet in release notes. See today's lead. (our story)
• Also this week: 1Password sign-in for Grok Bot (read) and X Spaces and Lists through the X connector (read); Bugbot stalled on Bitbucket (read); Claude API version 2026-10-21 drops temperature (read); OpenAI's approval rule for routers and AI gateways from Nov 20 (read); and Adcock says one Hark day can use a fifth of your month (read).

Opinion: Claude platforms and resellers should poll for bans for now. Anthropic now shows platforms when it restricts one of their end customers, but the webhook event isn't sent yet, so our top pick is a small poller. Second, map every route to OpenAI before the Nov 20 approval rule. Third, a sign-in check for Hark. Read the column

Daily Bot Check

1. Per Grok Bot's Oct 22 changelog, what can you do with a Page your Bot builds? (a) Only download it as a PDF (b) Edit it together or have the Bot keep it up to date (c) Post it to X automatically
2. As of Friday morning, where is the phone apps feature confirmed? (a) In @bot's posts only, not yet in store notes, the changelog or docs (b) In the App Store release notes for 1.17.0 (c) In the Grok Bot changelog for v0.75.2
3. When does Anthropic say auto mode becomes the default in most Claude apps? (a) It already is, since August (b) Next week, with no day given (c) On Nov 20

Answers: 1 (b), 2 (a), 3 (b).

Every story in this email

• Grok Bot can now use Calendar, Reminders and Contacts on your phone
• Grok Bot 0.75.2 notes: Bots can make videos and pay with Shop Pay
• Most Claude apps will start new sessions in auto mode next week
• Codex CLI 0.164 saves prompts in ten numbered preset slots
• Claude API run by Anthropic is now available through Google Cloud
• Open Responses moves to the Agentic AI Foundation
• What METR and Redwood found in OpenAI's agent monitoring
• Crypto guides tell users to run Binance's wallet skill in Hark
• Grok Bot can now sign in for you through 1Password
• Bot computer updates on Oct 17, 21 and 22 dropped installed apps
• Bugbot stalled on Bitbucket, and GitLab automations won't start
• OpenAI's Decisions API is now generally available
• Grok Bot for trades offices: receipts filed to jobs twice daily
• Claude platforms and resellers should poll for bans for now
• Grok Bot starts rolling out pages for reports, memos and dashboards
• Full issue, Oct 23

You're getting this because you subscribed at thedailybot.vercel.app. Reply with corrections or tips; a person reads every reply.

How we label: confirmed means we checked a primary source or a staff post; reported means a credible claim we haven't verified; "Our read" is our own analysis.