 |
| TUESDAY, NOVEMBER 3, 2026 · EDITION #345 |
|
START HERE
Build a channel around what you know
A childhood memory of HBO and Showtime became The Manhattan Network. Today’s workflow connects that story to a business channel built around your expertise, with Xmind as a tool for organizing the plan.
First, the news: Kittl and Synthesia bring creative work into chat, stolen AI access produces costly cloud bills, Google adds custom video presenters, and OpenAI proposes a way to ease pressure on local electricity grids. There are 26 stories in all.
|
|
|
01 · TOP FIVE
2026-11-03
Kittl’s new connections let people create product mockups, social images and video inside ChatGPT or Claude. A saved Brand Kit supplies the colors, logos and other choices that should stay consistent across a campaign. The tools also cover repeated jobs such as background removal and upscaling.
For a small retailer, the useful test is whether one approved product image and a clear brief can produce several usable pieces without rebuilding the brand instructions each time. These are Kittl’s announced capabilities, not our hands-on results. Its limited image promotion applies to paid plans and should not be mistaken for unlimited free use.
Kittl
|
|
|
02 · TOP FIVE
2026-11-03
Synthesia’s new ChatGPT plugin turns prompts, documents, presentations and existing conversations into videos, then lets users request changes in the same chat. Its Syren system handles generation, while Studio supports exact scripts, templates and more detailed editing.
That could make an existing onboarding document or customer explanation a useful starting point for a video. A person still needs to check what the finished presenter says and shows. The plugin is available across Synthesia plans, but video generation and chat edits consume credits. Basic-plan output includes a watermark.
Synthesia
|
|
|
03 · TOP FIVE
2026-11-02
Google’s Mandiant investigators describe criminals stealing organizations’ access to AI models and reselling that access to other users. In the incidents they investigated, unauthorized usage could exceed $100,000 in a day. Some victims accumulated roughly $1 million before access was revoked.
Those figures describe particular incidents, not the typical bill from an AI account. The practical check is whether your business knows where its access keys are stored, which services can spend money, and what alerts or limits are in place. Unusual new services and sudden usage increases deserve attention.
Google / Mandiant
|
|
|
04 · TOP FIVE
2026-11-02
Google Vids can create a custom video avatar from an uploaded image, with options to adjust its lighting and background and reuse it within the project. The feature gives people another way to make explainers and staff-training videos without repeatedly recording a presenter.
Availability is staged. Rapid Release rollout began October 30 and can take up to 15 days; Scheduled Release begins November 20. The announcement does not mean every customer has it today, and it should not be read as a promise to clone someone’s voice automatically.
Google Workspace
|
|
|
05 · TOP FIVE
2026-11-02
OpenAI’s Grid Protectors program is a plan for future computing campuses to reduce electricity demand when local grids are under strain. Its planned Effingham County, Georgia, campus includes an agreement for up to 1,000 megawatts of flexible demand response.
The idea matters beyond AI companies: households and small businesses need dependable electricity during extreme heat and cold. OpenAI says it will pay the electricity and infrastructure costs its campuses require. The Georgia facility is still in development, and the wider arrangements will vary by location. This is a commitment to future flexibility, not evidence that the program has already lowered household bills.
OpenAI
|
|
|
AI WORKFLOWS · TUESDAY
Context and harness management: keep the material, instructions and controls for an ongoing job together.
Don’t have it? Make your own.
When I was a kid, HBO and Showtime were status symbols.
I wanted the movies, of course. But having those channels also meant something to me. If we had them, maybe I wouldn’t feel as poor as I did.
Whenever a free preview weekend came along, we’d tape the entire thing, or close to it. Those tapes were like gold. HBO and Showtime were always on their best behavior during those weekends, showing the good stuff to convince people to subscribe.
I could replay those tapes and pretend we actually had the channels.
That memory stuck around. So did the bumpers, the feature presentation intros, and the feeling that something was always on, whether you were watching or not.
Recently, after seeing what other people were making with AI, I realized I had the tools to build my own version.
I called it The Manhattan Network.
It has programming, a guide showing what’s coming next, and original bumpers between shows. I’ve been working toward the look and feel of early-’80s premium cable, right down to the feature presentation and content advisory screens.
The first time it worked, I felt capable of something that had seemed unattainable in childhood.
Don’t have it? You can just make your own.
I didn’t begin with a perfectly mapped-out channel. At first, I wanted to see what content was available and whether the whole thing could work. When it pulled material from the Internet Archive, my next thought was: Why couldn’t it also use YouTube and Vimeo, where newer material already lives?
That opened up more possibilities. Feature films. Fan films. Concerts. Classic sports. Shorts. Even videos about making things with AI.
I baby-stepped it into existence, then made it better through testing and iteration. Watch it. Notice what feels wrong. Explain what needs to change. Try again.
Some of that has been as specific as telling the AI that THE MANHATTAN NETWORK should be more prominent than NEXT ON. It’s my network, after all.
And no, everything available is not going to be worth watching. Choosing what belongs is part of the work.
That’s where curation and creation meet.
HBO and Showtime built identities partly around programming made by other people. The selection mattered, but so did how they presented it and what viewers came to expect.
A business can use a similar approach to build a channel around its expertise.
A garden center could combine its own planting demonstrations with carefully selected videos about seasonal care. A fitness coach could organize workouts, technique demonstrations and answers to questions clients keep asking. Each could add original introductions explaining why the material matters.
There’s a reason to consider it. In Wyzowl’s 2026 video marketing survey, 63% of consumer respondents preferred learning about a product or service through a short video, compared with 12% who preferred an article.
Your channel doesn’t have to run 24 hours a day. Start with a useful collection organized around something your customers want to understand. Use AI to help research material, build the guide, create introductions and connect your own work with the pieces you’ve selected.
Then watch it yourself. What’s missing? What feels repetitive? Where does your experience add something nobody else has said?
Your judgment gives people a reason to come back. AI helps you put that judgment to work in a form you might not have been able to build before.
For me, that started with some treasured videotapes.
For your business, it might start with a question you answer every week.
|
Give your channel a standing brief.
For Tuesday’s skill, keep the channel’s working instructions in one place: who it serves, the questions it answers, your tone, approved sources, and what does not belong. ChatGPT Projects can keep related instructions, files and conversations together for repeated work, subject to your plan and workspace settings.
Try one small collection for the garden center example. Ask the AI to research five candidate videos from your approved sources and list each link, topic, length and reason it fits. Have it flag anything it cannot verify, draft two original introductions and suggest a viewing order. Review the choices, timing and introductions before anything is published.
Save your decisions and corrections with the brief. On the next pass, have the AI use them to improve the collection. Your finished result is a reviewed lineup and a reusable set of instructions for the next one.
|
|
|
|
The rest of today’s headlines
More tools, research and decisions worth knowing about.
|
|
06 · TOOLS AND BUSINESS
2026-11-02
Aion’s new finance agent connects work across its banking and payments platform, accounting systems and customer or vendor portals. It can help with invoices, reconciliation and reporting. The service requires Aion’s platform, and its approval controls can include bounded standing instructions. Start by examining the reporting work it can do before giving it authority over money. Aion is a fintech company, not a bank.
Aion · Supporting source
|
|
|
07 · TOOLS AND BUSINESS
2026-11-02
Sequel’s newly available agents help plan webinars, repurpose recordings into clips and recaps, and prepare follow-ups based on attendee activity. Its announcement says sending messages and transferring leads into customer-management systems require human approval. The tools are available to Sequel customers and are free for now; usage-based pricing starts in March 2027.
Sequel
|
|
|
08 · TOOLS AND BUSINESS
2026-11-02
Hera’s new creative board combines images, designs and animation while keeping text, logos and shapes in separate editable layers. That addresses a familiar problem: changing one detail after an AI has produced the whole design. It is an early release. The cofounder describes the first version as rough and says detailed manual animation controls remain limited.
Hera · Supporting source
|
|
|
09 · TOOLS AND BUSINESS
2026-11-03
Google Meet administrators can display a short policy message and link before people join a call. The notice can appear when recording, transcription or Gemini notes are active. It is off by default and rolling out on eligible organizational plans. It gives companies a place to explain their rules, but does not itself establish consent or guarantee legal compliance.
Google Workspace
|
|
|
10 · TOOLS AND BUSINESS
2026-11-02
Salesforce signed an agreement to acquire Cresta, whose products include AI voice agents, coaching for human representatives and conversation analysis. Salesforce plans to connect those capabilities with Service Cloud. The deal has not closed; the company expects completion in its first quarter of fiscal 2028, subject to approvals. Its announcement does not disclose a price.
Salesforce
|
|
|
11 · TRUST AND RESEARCH
2026-11-02
WorkOS disclosed unauthorized access to an internal application containing sales and support information. Exposed categories included business contacts, account histories and excerpts or summaries of customer conversations. WorkOS identified AI-assisted reconnaissance and exploitation. It says there is no evidence that production services or application data were accessed. The investigation is ongoing, and convincing requests that reference real conversations still need verification.
WorkOS
|
|
|
12 · TRUST AND RESEARCH
2026-11-03
CrowdStrike announced controls for interrupting unauthorized AI activity, alongside agreements to acquire monitoring company Fiddler and AI-security testing company Dreadnode. Initial enforcement operates through its gateway for AI access; controls that act directly on individual computers are still coming. Both acquisitions remain pending. Businesses should check which controls are available now before relying on the broader announcement.
CrowdStrike · Current product scope · Fiddler agreement · Dreadnode agreement
|
|
|
13 · TRUST AND RESEARCH
2026-11-02
The New York Times filed a complaint against Brave, while the Daily News, Chicago Tribune and Orlando Sentinel filed a separate case. The newspapers challenge alleged scraping and distribution of their journalism through search, AI answers and Brave’s developer service, including alleged paywall bypass. These are new allegations, not court findings. The cases matter to publishers and businesses building products around third-party material.
New York Times complaint · Other newspapers’ complaint
|
|
|
14 · TRUST AND RESEARCH
2026-10-31
xAI’s documentation says seven supported models will apply invisible text watermarks by default starting November 7. The change covers its developer service, including ordinary text and structured output. Administrators can opt out for some models, while others may require it. Detection is statistical and can be uncertain. The notice does not establish the same policy for the consumer Grok app.
xAI documentation
|
|
|
15 · TRUST AND RESEARCH
2026-11-02
Stanford researchers revisited 50 widely cited AI findings across 20 models. Under their criteria, 44% had largely faded in newer models, while other problems, including user flattery and favoring a model’s own writing, persisted. It is a selected set of findings, not a sample of all AI research. Before relying on an old conclusion, check which model and settings the study tested.
Original research
|
|
|
16 · TRUST AND RESEARCH
2026-11-02
Anthropic’s FlashProve prototype uses cryptographic proofs to verify that a provider ran the promised AI model without exposing its internal settings. That could help buyers confirm what they are paying for. The research system has substantial computing overhead and supports limited settings. Verifying which model produced an answer does not establish that the answer is correct.
Anthropic
|
|
|
17 · EVERYDAY USE AND WORK
2026-10-31
Andrej Karpathy’s HackerTube organizes 6,454 YouTube videos into 150 topics using recommendations gathered from Hacker News. It displays frames from the videos instead of promotional thumbnails and keeps recommendations locally without accounts. The site is a consumer discovery tool and an example of making software around personal taste. Karpathy’s account that he built it in about two hours still involved technical setup.
Karpathy’s post · HackerTube
|
|
|
18 · EVERYDAY USE AND WORK
2026-11-03
Google announced new U.S. teacher-training modules arriving November 4. They cover interactive assessments, game-based AI learning and strengthening assignments against shortcuts. The training is free. Expansion into additional countries and regions is planned for later months, so today’s announcement should not be read as immediate worldwide availability.
Google
|
|
|
19 · EVERYDAY USE AND WORK
2026-11-02
Zutec’s new Handover AI names, tags and classifies construction documents, identifies missing submissions and creates a building record people can question with links to the sources. It is available now and primarily serves UK and Ireland construction workflows. Human document checks remain part of the service; the announcement does not establish automatic legal compliance.
Zutec
|
|
|
20 · EVERYDAY USE AND WORK
2026-11-01
Proton CTO Bart Butler published an engineering manifesto arguing that faster coding can simply move delays into testing, infrastructure, approvals and reviews. His proposal is to change the surrounding work as AI speeds up production. It is a useful management question for small teams: what starts waiting when one step gets faster? The post is executive opinion and company strategy, not a measured productivity study.
Proton
|
|
|
21 · BEHIND THE TOOLS
2026-11-02
Claude’s new message threads let developer applications send a conversation once and then transmit only new material. Anthropic reports about 35% less waiting before output begins on very long Claude Code requests. The feature is in public beta. Claude still reads the full conversation, so token usage and standard rates remain unchanged; the result is not a claim that every task finishes 35% faster.
Anthropic
|
|
|
22 · BEHIND THE TOOLS
2026-11-02
OpenAI published an account of building its new code-storage system with Codex in ten weeks. One engineer directed the work of coding agents, with other employees helping deploy it. The system began production use in April. This is a newly published internal case study, not a new public product or an independently measured productivity benchmark.
OpenAI
|
|
|
23 · BEHIND THE TOOLS
2026-11-02
OpenAI increased its rewards for serious vulnerabilities, with the critical tier reaching a base maximum of $30,000 before bonuses. It also announced another internal security-testing effort. Larger exceptional awards are possible, but they are not the standard payment for every critical bug. The change gives outside researchers a stronger financial reason to report faults responsibly.
OpenAI
|
|
|
24 · BEHIND THE TOOLS
2026-11-02
MIT researchers had GPT-6 Astra edit a small portion of another model’s internal numerical settings to address reasoning loops. On the reported math test, accuracy rose from 7.9% to 48.1% under the same sampling settings. These are early experiments addressing a specific known failure in a small model, not independently replicated evidence of general AI self-improvement.
MIT researchers
|
|
|
25 · BEHIND THE TOOLS
2026-11-03
Britain’s National Cyber Security Centre warns that AI-assisted ransomware could threaten critical infrastructure by 2028. Its assessment says AI can accelerate the discovery and exploitation of existing weaknesses. This is a UK-focused future-risk assessment, not a report that such an attack has already occurred. The business lesson is to address known security gaps as attackers’ tools become faster.
NCSC
|
|
|
26 · BEHIND THE TOOLS
2026-11-02
NVIDIA released related open-source tools for diagnosing and recovering failed AI training jobs. Attribution combines logs and system information to identify likely causes; Recovery can isolate faulty equipment and restart from saved progress. New Recovery installations begin in observation mode so operators can check recommendations before allowing automatic actions. Some platform support remains on the roadmap.
Attribution · Recovery
|
|
mAIn Street · Aric Mitchell
Research dated November 3, 2026.
Availability and prices can change. Product descriptions are based on announcements and documentation, not firsthand tests. |