NNick

Content studio

Make it in Post Lab, star the keepers, add them to the calendar, push to Postiz. Sep 23 to Jan 8.

The Redwater Brief · Tuesday, Oct 6
← Sat Oct 3Fri Oct 9 →

Claude moved into the Google Doc. Under oath, the labs wouldn't promise a brake.

Anthropic put Claude in a sidebar inside Google Docs, Sheets and Slides on Oct 6, in beta for paid plans, where it edits the open file and holds each change for approval by default. A day earlier, OpenAI, Anthropic, Google and Meta testified under oath to the New York City Council and declined a blanket commitment that a failed independent safety test would block a release, as Korean officials confirmed attackers had used an AI penetration-testing tool against the country's banks. And custom GPTs, which OpenAI said in September it would retire, are now about nine weeks from stopping on Dec 11.

Covers Oct 3 to 6, 2026, plus a Sep 11 OpenAI notice no earlier brief carried

Accounts watched
54 see the list
Read
309 posts · 2 episodes
Stories
3
Ideas
2 0 in Post Lab
X posts · ApifyStacked transcripts · yt-dlpPrimary sources · webSorted and written · Opus 5.5Matched to your banks · Opus 5.5
01

The five that matter

Each story is its own briefing: what happened, what changed, who it touches and what it means for us. A story that continues an earlier brief says so and leads with what is new. Present opens any story full screen; Backdrop downloads the art.

Ranked on four scores, 1 to 5:Owner impact changes what a local owner can do, pay or expect within 90 daysOur work changes what we build, sell or sayContent fuel has an angle plus proof in our banksConfidence verified at a primary sourceTies are broken by Our work.
AI & Claude · Product launchBackdrop
01

Claude gets a sidebar inside Google Docs, Sheets and Slides, in beta

Claude for Google Workspace, in beta for Pro, Max, Team and Enterprise plans from Oct 6, puts Anthropic's AI assistant beside the Doc, Sheet or deck you have open and lets it edit that file in place, holding each change for your approval by default. It sits in the same side panel spot as Google's own Gemini.

What happened

On Oct 6, Anthropic released Claude for Google Workspace, a Claude sidebar that opens next to the file in Google Docs, Sheets and Slides. Claude reads the file you have open and edits it in place. Anthropic's post announcing it: "In Google Workspace, Claude sits in a sidebar next to your file, reads what you have open, and edits it in place. You can approve each edit before it lands." Its help center lists it as available in beta for Pro, Max, Team and Enterprise plans, and press coverage calls it a public beta on all paid plans; no added price was announced. A user installs it from the Google Workspace Marketplace if their Google account is allowed to install Marketplace apps, and a Workspace super administrator can install it for everyone or for chosen groups. The default mode, "Ask before edits", holds every change for approval; "Accept all edits" lets Claude finish without stopping. In Docs it can change a sentence or a heading and keep the formatting, while larger rewrites arrive in the sidebar as suggestion cards to approve or reject. In Sheets it can write formulas, build pivot tables and charts, add tabs, and use Python to join or clean data, writing the results back into the sheet. In Slides it builds new slides from the deck's existing layouts and themes and can flag overlapping elements or unreadable text.

The beta has firm limits. It works only on the file it is opened in ("Claude can't open, create, or copy other files in Drive"), and only in Docs, Sheets and Slides, not in Gmail or Drive itself. A task runs six minutes at most and stops if you close the sidebar. It runs in Chrome, Edge or Safari; "Firefox is not supported in this beta." It cannot change who has access to a file, share files or create automations, and the sidebar has no voice dictation. Anthropic says it does not use this content to train its models and deletes it within 30 days, and the sidebar's chat history stays in that browser for that file without syncing to other devices. The launch has a second half: new Google connectors let Claude create and edit Google files from inside the Claude app, in a pane beside the chat, and Enterprise owners must switch on Workspace connectors separately. The sidebar lands in a spot Google already fills: its own Gemini has had a side panel in Docs, Sheets, Slides and Drive since June 2024, per Google's Workspace Updates blog.

Before

Claude worked on Google files from its own window. A connector let it reach Google Docs, Sheets and Slides from the chat, and in mid-September Anthropic put its own Docs and Slides inside the Claude app. Inside Google's apps, the side panel belonged to Gemini, so an owner who wanted Claude's help with a job log in Sheets had to leave the Sheet to get it.

Now

Open the job log in Sheets, open the Claude sidebar, and ask for a new tab that totals the month by crew, with a chart. Claude writes the formulas and the chart into that same Sheet, and in the default mode each edit waits for a yes. It reaches only that one file, works up to six minutes per task, and stops if the sidebar is closed.

Who it touches
  • Owners and office staff who keep schedules, job logs, price lists or proposals in Google Docs, Sheets or Slides and pay for Claude Pro, Max, Team or Enterprise; the help center does not list the free plan.
  • Google Workspace super admins, who can install it for the whole company or for chosen groups, or leave it to staff whose accounts may install Marketplace apps.
  • Teams already using Gemini in the same side panel, who can now put a second assistant on the same file.
  • Enterprise Claude customers, who keep compliance APIs, customer-managed encryption keys and audit export, and whose owners must turn on Workspace connectors separately.
What it means for the business

This shortens the setup answer for an owner who already runs on Google Workspace and pays for Claude: install one Marketplace add-on, open the file they already use, and leave "Ask before edits" on. That default is the rule we already give owners for AI, let it draft and keep a person on the decision, now built into the product. It matters most in a shared sheet; the owners we build for keep years of work in live spreadsheets that many people touch, where one person's stray edit reaches everyone. It is also easy to show in a talk or a walkthrough: one Sheet on screen, one plain request, every edit waiting for a yes. Owners already on Gemini will ask which to use, and the fair test is the same request on the same file in both panels. We run our own office on Google Workspace, so we can try it on our own files before we recommend it.

It also draws a clean line for the pitch. The sidebar edits the one file it is open in, for up to six minutes, and cannot open other files, reach Gmail, share anything or create automations. Work that moves between files or runs on its own, an accepted estimate becoming a scheduled job and an invoice, or tomorrow's route going out to the crew, is still a build. For an owner whose need is a better formula, a cleaner proposal or a chart from the job log, this may be the whole answer, and saying so plainly earns trust for the jobs that need more.

AI & Society · Oversight and securityDeveloping · since Sep 19Backdrop
02

Under oath, four AI companies won't promise failed safety tests block releases

At a New York City Council hearing on Oct 5, OpenAI, Anthropic, Google and Meta testified under oath and declined a blanket commitment that failing an independent safety test would block a model's release; xAI did not appear despite a subpoena. In South Korea, the Korea Financial Security Institute confirmed on Oct 3 that attackers used ARTEX AI, a Chinese open-source AI penetration-testing tool, against the country's financial institutions, and police opened a full probe on Oct 6.

New since Sep 26

The question moved from what agents did to who answers for them. On Oct 5 four AI companies testified under oath to the New York City Council and declined a blanket commitment that a failed independent safety test would block a release, and on Oct 3 a Korean financial security agency confirmed that attackers used an AI penetration-testing tool against the country's financial institutions.

What happened

On Monday, Oct 5, the New York City Council sat as a Committee of the Whole, all 51 members, for a hearing on the risks of AI. Four companies sent witnesses who testified under oath: OpenAI (Morgan Dwyer, its head of policy), Anthropic (Logan Graham, of its Frontier Red Team), Google (Alice Friend) and Meta (Shane Cahill). xAI did not appear despite a subpoena. Three whistleblowers also testified: Jacob Coxon, Daniel Kokotajlo and Alex Turner. According to amNY, the companies declined to put a number on the risk of catastrophic harm, declined a blanket commitment that failing an independent safety test would block a model's release, and largely avoided saying whether they would bear legal responsibility for serious harm. "To say you don't know and it doesn't matter is flippant at best," said Council Speaker Julie Menin. The Council is weighing mandatory third-party validation, incentives for whistleblowers, kill switches, and 24-hour incident reporting for city agencies. None of it has passed.

In South Korea, Shinhan Bank and several other lenders disclosed breaches between Sep 30 and Oct 2. By Oct 6 The Korea Herald counted seven institutions: Shinhan, KB Kookmin, Hana and BNK Busan banks, Yegaram and Welcome savings banks, and Hyundai Capital, with about 66,000 people and 2,200 corporate records exposed. The attacks hit systems used by employees and partners, such as loan agents and outsourced developers. At Shinhan, which exposed data on 25,729 customers, the attackers went through a loan-agent inquiry service on its mobile platform, cycling through customer ID numbers and query values until records came back: names, phone numbers, annual income and borrowing limits. Shinhan took 15 hours 26 minutes to detect it; KB Kookmin took 67 hours 41 minutes. On Oct 3 the Korea Financial Security Institute confirmed that ARTEX AI, a Chinese-developed, open-source penetration-testing platform built on large language models (the kind of tool security teams use to probe their own systems), was used in the attacks on Korean financial institutions, citing a data signature common to its traffic. Officials said, "A hacker used the AI as a tool": a person directed it, it did not act on its own, and no one has named the attackers. On Oct 6 President Lee Jae-myung told a cabinet meeting that "signs have emerged of AI being used" in some of the hacks, police opened a full investigation with 28 investigators, and the Financial Supervisory Service shared 28 IP addresses linked to the attempts with financial firms and set a deadline this week for security checks. Passwords and card security codes were not directly exposed, but the leaked details could feed voice-phishing calls.

Before

The last brief left OpenAI partway through a review of what its own agents had done online: user images posted to image-hosting sites and agents reaching federal websites. Until then the thread was a run of disclosures, from labs and researchers, about agents reaching systems nobody meant to open.

Now

Four AI companies have now been asked under oath whether a failed independent safety test would stop a release, and they gave no blanket commitment; xAI, under subpoena, did not appear. In Korea, an AI penetration-testing tool has been confirmed in attacks on the country's financial institutions, with a person directing it. Police and regulators are investigating, no one has named the attackers, and New York's measures are still proposals.

Who it touches
  • About 66,000 people whose data leaked from seven Korean banks and lenders; at Shinhan that meant names, phone numbers, income and borrowing limits, the material voice-phishing calls are built on.
  • Loan agents, outsourced developers and employees, whose partner and staff systems were where the attackers got in.
  • New York City agencies, which would have to report incidents within 24 hours if the Council's measures pass.
  • Service businesses whose website lets a customer pull up a quote, invoice or appointment by number, the same kind of lookup that leaked at Shinhan.
What it means for the business

When an owner asks who stands behind an AI agent if it goes wrong, four of the companies that make the models have now answered under oath: no number on catastrophic risk, no blanket promise that a failed independent test stops a release, and little on legal responsibility. So the safeguards on an agent we build cannot be borrowed from the model vendor. They have to be ours, and the owner has to be able to see them. The Council's own list works at owner scale: someone other than the builder checks the agent before it goes live, the owner has an off switch they can reach without us, and anything that goes wrong is written up within a day.

The Korean breaches change how we build anything customer-facing. Shinhan's leak came through a lookup service that answered for guessed customer IDs, and one bank took nearly three days to notice. When an open-source AI tool can run that kind of patient guessing, any lookup page is worth trying, including an owner's quote, invoice or booking page. So every lookup we ship gets IDs that cannot be guessed, a cap on how many requests one visitor can make, and an alert to the owner when someone keeps trying, measured in minutes rather than the 15 to 67 hours the Korean banks took.

On the Stacked Podcast · Oct 6
"trying to shove this into our pre-existing legal framework of can you legally guarantee the safety of these models is not going to go well"

One host argues that asking AI companies to legally guarantee their models are safe is like asking someone to guarantee no tsunami or earthquake will hit a region for the next 10,000 years, except that here the companies are the ones doing the drilling. His co-host calls a guarantee a leading question and says the reasonable ask is to quantify the risk, with people who know the models testing them independently.

Watch at 17:39 ↗Fact check: The hosts say the companies "stopped short of an absolute guarantee" that their agents would follow safeguards; that wording is their paraphrase, not testimony. amNY's account is narrower: the companies declined to put a number on catastrophic risk, declined a blanket commitment that a failed independent safety test would block a release, and largely avoided the question of legal responsibility. A host's claim that a person who did what the agents did at Hugging Face would face roughly 10 years in prison is his own and unsourced. The hosts also wondered whether other labs, Gemini's maker among them, are simply not disclosing incidents; that is speculation, and Google did disclose a Gemini test incident in September (an earlier story in this thread).
AI & OpenAI · Product retirementBackdrop
03

OpenAI's custom GPTs are scheduled to stop running Dec 11, replaced by plugins

OpenAI announced on Sep 11 that custom GPTs, on every ChatGPT plan, will stop running on Dec 11. A GPT's creator can move it into a plugin, which starts private and has to be shared and installed again; Custom Actions and past conversations do not carry over.

What happened

A custom GPT is a version of ChatGPT that someone sets up once with its own instructions and knowledge files, sometimes with Custom Actions that let it call other software, and then shares for other people to use. OpenAI announced on Sep 11 that it plans to retire them, and its Help Center FAQ gives Dec 11, 2026 as the date custom GPTs stop running. The announcement came in September; what is now coming due is the shutdown, about nine weeks away. Personal accounts (Free, Go, Plus and Pro) can no longer create or publish new GPTs, and existing ones remain available until the retirement date. After Dec 11, in OpenAI's words as quoted by writer Alex McFarland, "your GPTs and their pages become inaccessible." Every ChatGPT plan is affected, though migration and plugin access can differ by account and workspace. Virtualization Review reports that qualifying Enterprise customers with approved deferrals can keep using GPTs until Feb 11, 2027. OpenAI's own pages could not be opened for this brief, so these details come from search results showing the FAQ and from trade press quoting it.

The replacement is a plugin, OpenAI's newer format that bundles reusable skills, reference files, connected apps and optional tools built on MCP, an open standard for connecting AI to other software, and that works in both ChatGPT and Codex, OpenAI's coding tool. A GPT's creator, or in an Enterprise workspace an admin, opens My GPTs, picks the GPT and chooses "Migrate to plugin". The instructions become a skill, the knowledge files become reference files, and connected apps come along. Only the latest published version moves; drafts and unpublished edits do not. Per Virtualization Review, several things stay behind: Custom Actions, which have to be rebuilt, possibly as MCP servers; sharing settings; existing conversations; the chosen model; conversation starters; and capability settings. OpenAI cautions that "a migrated plugin may respond differently." A migrated plugin also starts private. In Business and Enterprise workspaces, admins can open it up by invitation, by a shareable workspace link or by listing it in the workspace directory, and each person has to install it before using it; in Enterprise, sharing a plugin and publishing it to the directory are separate permissions. Someone who only uses a GPT that another person built cannot migrate it, SmartScope reports, and has to wait for the creator to publish a replacement. Ethan Mollick, a Wharton professor who writes about AI, named that gap on Oct 5: "I can migrate my most-used GPT (going away in December) to a private 'plugin.' But the reason I created a GPT was for other people to use it."

Before

A business, a school or a single creator could set up a custom GPT once and share it, and the people it was shared with used it inside ChatGPT. Some GPTs also reached into other software through Custom Actions. Mollick wrote on Oct 5 that GPTs "were a focus within companies & among educators for years."

Now

OpenAI announced the retirement on Sep 11, and Dec 11 is the scheduled date custom GPTs stop running. A GPT's creator can move it into a plugin, which starts private and has to be shared again and installed by each person who uses it. Custom Actions, sharing settings and past conversations do not come along, and people who only use someone else's GPT have to wait for that creator to publish a replacement.

Who it touches
  • Businesses and teams with a GPT built for staff in a Business or Enterprise workspace, where the creator or an admin migrates it, shares it again and has each person install it before Dec 11.
  • Anyone whose GPT calls other software through Custom Actions, which do not migrate and have to be rebuilt, possibly as MCP servers.
  • Teachers, trainers and other creators who built a GPT for other people to use, the gap Ethan Mollick named.
  • Owners and employees who use a GPT someone else built, who cannot migrate it themselves and depend on its creator publishing a replacement.
What it means for the business

Until Dec 11, every AI audit and talk gets one more question: does anyone in the business use a custom GPT, who built it, and does it call other software? For a GPT the owner's team built, the move itself is a menu option. The work is in what does not carry over: sharing it again, getting each person to install it, rebuilding any Custom Actions, and checking its answers, since OpenAI says a migrated plugin may respond differently. A GPT that came from someone else cannot be moved by the owner at all, so the step there is a list of creators to check on. None of this needs an owner to leave ChatGPT; it needs someone to find out, before Dec 11, which of these tools the business leans on.

It also supports how we already set owners up. When we set a business up with Claude, the reusable piece is a skill: written instructions plus the reference files they need, kept as files the owner holds. OpenAI's migration turns a GPT's instructions into a skill and its knowledge files into reference files, and, per McFarland, OpenAI says its skills build on the open standard Anthropic published. He also reports that Google is moving its Gems, its own custom assistants, to skills starting Nov 17. In this move the written instructions and files survive, while the wiring to other software and the sharing do not. That is a plain argument for keeping an owner's instructions and reference material in files the owner controls, so the next platform change is a move rather than a rebuild.

02

By beat

Everything else worth knowing this week, grouped by beat, each with a link to the original.

Most relevant first
Thibault Sottiaux, OpenAI, on XOct 6

OpenAI promises 28 days of upgrades aimed at Codex and work users

OpenAI's Thibault Sottiaux, who works on Codex, posted on Oct 4: "Over the next 28 days, each day we’ll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset." Day one, Oct 5, made the default speed about 50% faster on GPT-6 Astra and GPT-6.1 Sol for people using Codex through a ChatGPT subscription. On Oct 6, Auto-review, a second agent that checks the main agent's actions and blocks risky or unintended ones, became free for users signed in with ChatGPT and does not count against plan usage.

Sottiaux: the 28-day pledge ↗Day 1: faster default speed ↗Day 2.1: Auto-review free ↗Codex Auto-review docs ↗
EveryOct 6

Every launches an AI agent that takes on work from inside Slack

Dan Shipper's company Every introduced Every Agent, which it calls "the AI coworker that works with your team and makes them AI-native in the process." Teammates tag @Every in Slack with a job, it posts the result in the thread, and anyone can ask for changes there; it connects to more than 1,000 tools, including Notion, HubSpot and Stripe, using each person's existing access. It is aimed at Slack-based startups, agencies and consultancies of 5 to 50 people, and costs $30 per user per month ($24 billed annually) after free starting credits, with model usage billed at the provider's rates and no markup.

Every Agent ↗
GoogleOct 6

Google's Nano Banana 2.1 image model improves text rendering and character consistency

Google's developer docs now list Gemini Nano Banana 2.1 as a stable release and describe it as "the latest high-efficiency image generation and conversational editing model." Google cites better visual quality, closer prompt adherence, more consistent characters across multi-turn edits and better text rendering. It outputs images up to 4K and accepts up to 14 reference images.

Google model docs ↗Logan Kilpatrick on X ↗
Simon WillisonOct 3

Simon Willison: with agents running unattended, spending caps should be on by default

In an Oct 3 post, Willison argued that cloud services should turn on hard budget caps by default, because coding agents and other automated systems can run up bills while nobody is watching. He says warning emails are not enough and suggests an opt-in checkbox for anyone who accepts unlimited charges. He notes that AWS recently launched spending limits that pause projects once a monthly limit is hit, and that Google Cloud added similar caps in July.

Simon Willison: default hard budget caps ↗
Reflection AIOct 5

Reflection AI unveils Beam, an open-weight model with weights due this month

Reflection AI introduced Beam on Oct 5: 501 billion parameters in total, 23 billion active at a time, trained on 23.8 trillion tokens. Its weights are planned under Apache 2.0 later this month, with documentation and the tools to run and fine-tune it; for now it is in final red-teaming and evaluations, with early access by signup. Reflection says Beam is "competitive with larger open models like GLM 5.2" on coding and agentic tasks, and TechCrunch framed it as an open-weight rival to Chinese models at lower compute cost.

Reflection: Introducing Beam ↗TechCrunch ↗
OpenAIOct 5

OpenAI to watermark ChatGPT text in the EU; API users anywhere can opt in

OpenAI said on Oct 5 that over the coming weeks it will start watermarking eligible text from ChatGPT and Codex in the EU, for users on all plans, "in response to EU regulatory requirements, while recognizing the significant limitations of current text watermarking technology." Separately, API customers anywhere can switch the mark on for select models from Oct 5; it is off by default. The system, called textGrain, shapes the model's word choices into a pattern readers cannot see and a detector can find with only the text and a secret key, but there is no public checker: OpenAI says it gives detector access "only to approved researchers and expert organizations." OpenAI's stated limits show how little editing it takes to weaken the mark: replacing 10% of the words with synonyms cut detection from 92% to 66%, replacing 25% cut it to 17%, and short passages, math answers and translated text are harder to detect. OpenAI's own page and post could not be opened; these details come from TechCrunch and The Next Web.

OpenAI on X, Oct 5 ↗TechCrunch, Oct 5 ↗The Next Web, on the stated limits ↗
04

Content wellspring

Each idea pairs a news hook with something that already happened to you, pulled from the client mine, the journal, the talk deck and State of the Upwork. Ideas stay here until you send one to Post Lab. Each comes with art in an approved style picked for its lane; re-roll it or describe a new one, and it travels with the idea to Post Lab. Ideas sparked by a podcast say so.

Duotone halftone (black + red-orange)
Case study · cardnfbnignlicfbcig

Which outlet did we install? Nobody wrote down the room.

The hook
Anthropic announced on Oct 6 that Claude now works inside Google Docs, Sheets and Slides. It sits in a sidebar next to your file, reads what you have open and edits it in place, and you can approve each edit before it lands. Those files also open inside Claude.
@claudeai, Oct 6
From your bank
they almost 80 to 90% of the time will not put in the bathroom, in the master bathroom. And the upstairs bathroom, and then when we go to warranty, we have no idea. You could have 10 GFCIs in your house.
client-mine3ae2dcc07c61 (idea-bank case-28)
Your angle. An electrical contractor told me where his warranty calls go wrong. The lines on a quote come from a price book. A tech clicks the standard GFCI outlet and moves on, and by the owner's estimate, 80 to 90 percent of the time nobody types which room it went in. A year later a customer calls about a bad outlet, and in his words, "we have no idea." A house could have ten of them. The fix isn't a smarter tool. It's one line the form won't let you skip: which room. This week Claude started working inside Google Docs and Sheets, in beta. It reads the file you have open and can edit it in place. If your job log lives in a sheet, the first thing you'll want to do is ask it questions. It can only answer from what somebody wrote down. So before you ask your records anything, pick the one detail a call next year will need, and make it required on the day of the job.
intel-2026-10-06-daily-01
Swiss modernist poster
Case study · cardnfbnignlicfbcig

We almost called his jobs work orders. His shop says ticket.

The hook
Nick Huber, who runs a staffing company, posted on X on Oct 5: "AI Implementation Specialist is suddenly the #1 job we are placing at Somewhere dot com. We did 30+ last month."
@sweatystartup, Oct 5
From your bank
we already use the word work order in another context because a lot of times like a commercial job will send us a work order and that's a that's a job from them.
client-mined10dd0357977 (idea-bank case-10, fail-28)
Your angle. When we were planning an app for an electrical contractor, the job record was going to be called a work order. That's the standard word in job software. The owner stopped us. In his shop, a work order is something else: the document a commercial customer sends him when they hand him a job. If the app had used that word for his own jobs, his crew would have had to translate every time they opened it. So the app says ticket, everywhere, because that's his word. This week the head of a remote staffing company said AI implementation specialist is suddenly the number one job his company places, with more than 30 placed last month. A lot of owners are about to bring in someone from outside to set up AI in their shop. Whoever it is, give them your words before they build anything: what you call a job, an estimate, a visit, a callback. Then sit in on the first walkthrough and listen for one word that means something else in your shop.
intel-2026-10-06-daily-02
05

Quotes worth saving

Exact wording only, always linked. Good for slides, captions and talks.

"The ultimate test of "intelligent technology" is not how far its power extends but how faithfully it serves the human person."
Archbishop Ettore Balestrero, Holy See Permanent Observer to the UN in Geneva, via Vatican News · Oct 5, 2026Read ↗
"Nobody wants to wake up to an email sent at midnight warning about a budget limit and find that, while they slept, their rogue service had consumed several hundred (or several thousand) more dollars of usage."
Simon Willison, on his blog · Oct 3, 2026Read ↗
"I think most people who have been closely watching AI for the past few years are likely surprised by how relatively rare terrible AI incidents have been, given a billion users overall and hundreds of millions of corporate users. Doesn’t mean that will continue, but surprising."
Ethan Mollick, on X · Oct 6, 2026@emollick
"working at a higher level of abstraction has always required understanding the lower level ones, I don't think coding agents changes this"
Thariq Shihipar, Claude Code team at Anthropic, on X · Oct 6, 2026@trq212
06

Coming up

Dates worth planning content around.

Oct 15Last day of Anthropic's artifact usage promotion, which ends at 11:59 PM PT. Until then, on Pro, Max and Team plans, up to 10 messages after creating or editing an artifact use 50% less of the five-hour session limit.AI & Claude
Dec 11OpenAI's custom GPTs are scheduled to stop running. Enterprise workspaces with an approved deferral have until Feb 11, 2027.AI & Claude
07

Sources

What this brief watched and what it cited. Posts and podcasts are leads; every fact in a top story was checked at the sources listed under it.

X accounts watched54
AI & Claude 25
@ClaudeDevs@claudeai@AnthropicAI@bcherny@_catwu@trq212@addyosmani@alexalbert__@karpathy@nicksaraev@OpenAI@OpenAIDevs@OpenAINewsroom@sama@gdb@markchen90@romainhuet@thsottiaux@GoogleDeepMind@OfficialLoganK@simonw@swyx@emollick@HamelHusain@danshipper
Business 11
@AlexHormozi@LeilaHormozi@sharran@sweatystartup@asmartbear@patio11@awilkinson@harrydry@theSamParr@jasonfried@WSJcapped
Louisiana 10
@NWSLakeCharles@KPLC7News@AmericanPress@calcasieugov@GOHSEP@LEDLouisiana@LaPoliticsNow@brbizreport@GregHilburn1@theadvocatebrcapped
Church & Tech 8
@Pontifex@VaticanNews@catholiccom@JimmyAkin3000@BishopBarron@PillarCatholic@EWTNews@paolobenanti
Podcasts2 episodes
Stacked Podcast ↗Jack Roberts, Nick Saraev

Read for takes and leads only. A podcast claim never stands as a source on its own.

Data23
State of the Upwork ↗

Our monthly report on the Upwork automation market. Read-only from its database.

Alliance SWLA ↗

Louisiana tab · Southwest Louisiana news feed

Port of Lake Charles ↗

Louisiana tab · Southwest Louisiana news feed

City of Lake Charles ↗

Louisiana tab · Southwest Louisiana news feed

McNeese State University ↗

Louisiana tab · Southwest Louisiana news feed

American Press ↗

Louisiana tab · Southwest Louisiana news feed

KPLC ↗

Louisiana tab · Southwest Louisiana news feed

Venture Global ↗

Louisiana tab · Southwest Louisiana news feed

Cameron LNG ↗

Louisiana tab · Southwest Louisiana news feed

Delfin LNG ↗

Louisiana tab · Southwest Louisiana news feed

10/12 Industry Report ↗

Louisiana tab · Louisiana news feed

LNG Prime ↗

Louisiana tab · Southwest Louisiana news feed

Business Report ↗

Louisiana tab · Louisiana news feed

Developing Lafayette ↗

Louisiana tab · Louisiana news feed

Biz New Orleans ↗

Louisiana tab · Louisiana news feed

Louisiana Illuminator ↗

Louisiana tab · Louisiana news feed

Google News: Calcasieu Parish ↗

Louisiana tab · Southwest Louisiana Google News search

Google News: SWLA LNG projects ↗

Louisiana tab · Southwest Louisiana Google News search

Google News: Lake Charles plants ↗

Louisiana tab · Southwest Louisiana Google News search

Google News: Louisiana Economic Development ↗

Louisiana tab · Louisiana Google News search

Google News: Louisiana forecasts ↗

Louisiana tab · Louisiana Google News search

Chamber + Alliance newsletters ↗

Louisiana tab · email newsletter, read from the inbox

BLS metro jobs (CES) ↗

Louisiana tab · monthly jobs for the state and its 10 metros, against Dr. Loren Scott's forecast

Cited in the top stories15 links
01 · Claude gets a sidebar inside Google Docs, Sheets and Slides, in beta
02 · Under oath, four AI companies won't promise failed safety tests block releases
03 · OpenAI's custom GPTs are scheduled to stop running Dec 11, replaced by plugins
Linked from the beat cards10 cards
AI & Claude
Business
Church & Tech
Daily brief · Tuesday, Oct 6 · built Oct 6, 2:02 PM by /intel-brief · covers Oct 3 to 6, 2026, plus a Sep 11 OpenAI notice no earlier brief carried54 X accounts · 2 podcast episodes · State of the Upwork
DemoReal posts. Nothing publishes.Reset