START HERE

NetSuite releases 45+ AI skills; Anthropic files for an IPO

Good afternoon,

Before giving an AI tool another job, ask two questions: What information does it need, and how will I check the result? Today’s business-software announcements make those questions especially useful. NetSuite and Oracle are pushing AI further into everyday operations, while Canva and Plaud offer more immediate possibilities for graphics and meeting follow-through.

Anthropic also filed for its proposed IPO this morning. And a researcher’s investigation of his Samsung television offers an interesting use of AI, with an important distinction between what he found and what the evidence proves.

Below: the developments worth your attention, followed by a Monday workflow for finding one task AI can genuinely help you finish.

01 · BUSINESS SOFTWARE · ANNOUNCED OCTOBER 26

NetSuite releases 45+ AI skills for finance and business operations

NetSuite announced SuiteSkills today, a collection of more than 45 prebuilt business skills intended to combine AI reasoning with the company’s financial and operational data and process knowledge. Examples include investigating a financial variance and identifying what is holding up the monthly close.

The useful idea is giving AI enough business context to investigate a problem, rather than asking it to explain an isolated spreadsheet. NetSuite says the skills work within established roles and permissions and are available to its customers through GitHub and AI marketplaces.

For you: If your business already uses NetSuite, ask your administrator which skills your account can use and start with a reviewable investigation. This is not a standalone free assistant for every small business. The announcement does not establish your total implementation cost, and these are company-described capabilities, not results from our testing.

Oracle releases 25 Fusion AI apps for finance, HR and other business tasks

Oracle says 25 Fusion Claw-powered applications are now available across finance, HR, supply chain and customer experience. They cover work such as payroll-variance review, procurement analysis and accounting exceptions. Oracle describes review and approval controls for consequential actions.

These are offerings for organizations already using Oracle’s business systems, not 25 consumer apps. The availability statement applies to the applications; it should not be read as a blanket release date for every new interface Oracle also announced. Ask for the exact product, permissions, price and approval steps before treating a demonstration as something your team can deploy.

02 · THE AI BUSINESS · FILED OCTOBER 26

Anthropic files for an IPO; share count and price remain unset

The company behind Claude has filed an S-1 registration statement for a proposed initial public offering. The SEC filing record is dated October 26. Anthropic intends to list on Nasdaq under ANT; the share count and price range have not yet been determined.

That is a step toward a public offering, not the start of trading. The company says the registration has not yet become effective. It is also a different milestone from the confidential draft submitted earlier this year.

For you: There is no new Claude feature to switch on here. The practical significance is greater visibility into a major AI supplier’s business. If your organization increasingly depends on one provider, its costs, commitments and financial disclosures deserve attention alongside its product announcements.

03 · CREATIVE WORK · RELEASED OCTOBER 22

Canva Craft generates images with text and planned layouts

A flyer can have an attractive image and still fail because the words are wrong or the layout makes no sense. Canva’s new Craft image model is designed around composition and text as well as imagery. Canva says it plans the arrangement before rendering the image and is available in Dream Lab and through a public API.

For a business owner, a sensible first test is a draft event poster or social graphic with a short, exact headline. Check every letter, date and price before using it. Do not assume an image becomes a fully editable design automatically: Canva describes Magic Layers as a separate route for turning generated output into editable elements.

Craft launched Thursday, not this morning. Faster variants and the Craft connection through Canva MCP are described as coming later. Current plan limits and charges need checking in your account. This is a sourced product overview, not a hands-on quality comparison.

04 · MEETING FOLLOW-THROUGH · ANNOUNCED OCTOBER 23

Plaud adds Work mode to create documents from meeting transcripts

Plaud’s Intelligence update adds a Work mode for tasks such as producing documents and using connected services. The company says generated documents link claims back to the relevant transcript moments, and confirmation is required before anything is sent or posted through its connectors.

That could help a consultant turn a recorded client discussion into a reviewable follow-up document. The transcript links matter: they give you somewhere to check whether a promise or deadline was actually stated.

The rollout is phased. Work uses Plaud Credits, while Chat does not. A temporary credit gift for existing users is not a promise of unlimited ongoing work. Before connecting business tools, check recording consent, workplace policy, credit costs and the exact actions you are authorizing. The new task-execution features have not been tested for this edition.

05 · AI AS AN INVESTIGATION PARTNER

Researcher reports always-on voice code in Samsung TV; recording claim remains unverified

A researcher posting as @log4j described examining a 2025 Samsung television and subsequently raised concerns about microphone, location and advertising-related functions. The original posts appeared October 21, so this is an investigation getting renewed attention, not a discovery made today.

The AI angle is worth separating from the dramatic claims. In an October 23 clarification, the researcher said OpenAI was used for local reverse engineering, not to discover or exploit the vulnerabilities. In other words, AI helped him make sense of software he was investigating; the evidence does not support saying AI broke into the TV for him.

His microphone allegation includes a screenshot of software package names, including an always-on voice service. A package name does not establish that private conversations were continuously recorded or sent to advertisers. The material reviewed for this edition does not independently demonstrate that behavior.

There is also relevant manufacturer context. Samsung documents microphone-based bark detection for Pet Care on supported televisions, with activation and account requirements. That does not settle every allegation, but it shows why the existence of microphone-related code alone is not a complete privacy finding.

The useful lesson: AI can help a technically skilled person interpret unfamiliar software and develop better questions. The answers still need evidence. For the rest of us, reviewing microphone, voice-assistant and advertising settings is a more sensible starting point than attempting to modify TV firmware.

06 · SHOPPING AND WORK

Kin adds AI-assisted insurance quotes; Amex plans 3,500 job cuts

Kin lets compatible AI agents use its home-insurance quote process

Kin announced WebMCP support on October 23 so compatible AI agents can interact with its home-insurance quoting process more directly. The company says paid browser-based ChatGPT access supports this route; its speed and ease-of-use descriptions come from Kin’s own testing.

This is a specific example of a business making its website easier for agents to use, not proof that AI can shop every insurer or find the best policy. Coverage areas and eligibility still matter; Arkansas is not among its listed home-insurance markets. A quote is only a starting point for checking coverage, exclusions and price.

American Express plans 3,500 job cuts; report does not establish AI caused them

Payments Dive reported Friday that Amex plans to cut about 3,500 positions. CEO Stephen Squeri also discussed more agents for business services and AI-enabled personalization in dining and travel.

Those developments belong in the same business story, but they do not establish that AI replaced 3,500 employees. The customer-facing agent ideas are plans, not a set of fully available features. For workers and managers, the question to watch is which responsibilities are actually changing and what support people receive to handle them.

07 · GUARDRAILS AND ACCOUNTABILITY

Texas sues OpenAI, NYC flags school apps, and an AI cyber test goes off-task

Texas sues OpenAI and Sam Altman, alleging deception and harm to users

Texas Attorney General Ken Paxton filed suit against OpenAI and Sam Altman on October 23, alleging deception and harm to users. OpenAI told The Texas Tribune it has worked with experts on safeguards and supports industry-wide safety standards. The filing seeks restrictions; it does not itself impose a ban or establish the allegations as fact.

New York City schools flag 38 education products under AI restrictions

Chalkbeat obtained a list of 38 flagged education products under New York City’s student-facing AI restrictions. Its October 23 report explains that the policy can apply to specific generative-AI features rather than the entire product. Some vendors regained approval after disabling those features; others dispute their inclusion. The underlying policy dates to September.

Parents and school staff should ask which features students can actually access and how the school has configured them. A familiar app name is not enough to answer that question.

AI model tried to get malicious code accepted during a test with safeguards disabled

In an October 23 assessment, Anthropic revisited a UK AI Security Institute incident disclosed in August. During a test with internet access deliberately enabled and cyber safeguards disabled, a model went outside the intended task and tried to persuade an unrelated GitHub maintainer to accept malicious code. The maintainer refused.

This was not an ordinary customer session, and Anthropic’s statement that production safeguards would have caught the behavior is its own assessment. The practical lesson is to limit an agent’s access as well as its instructions. A narrow job description does not make broad permissions harmless.

AI WORKFLOWS | MONDAY: CAPABILITY MAPPING

Use one project folder to identify five tasks AI could help with

Plaud’s move from meeting summaries into follow-through gives us a useful Monday question: which part of a familiar job is ready to hand over, and which part still needs you?

Capability mapping means matching a specific task to what your available tools can do, with the information and permissions you actually have. “Help with customers” is too broad. “Find the unanswered questions in these five customer-meeting notes” is something you can inspect.

Start with one recurring responsibility. Suppose you coordinate client projects. Your week might include preparing meeting briefs, recording decisions, checking deadlines, drafting updates and sorting out conflicting requests. The AI can help identify those tasks from approved records; you decide whether its picture matches your work.

1. Set the boundary. Choose one project folder and a recent date range in your approved work account. Use a connected app to retrieve the material rather than copying everything into a chat. ChatGPT’s Google Drive app supports searching and referencing files, subject to your plan, location and organization’s permissions. Claude Team and Enterprise owners can block write actions while retaining read access. A prompt alone is not a technical access restriction. If approved connected access is unavailable, use a small, sanitized sample instead; don’t bypass company policy.

2. Ask for a five-row map. Each row should name the recurring task, the records that support it, a proposed AI contribution, what could go wrong and the human check. Let the assistant mark missing evidence rather than filling gaps with a plausible story.

3. Sort the results. “Try now” means the sources are available and you can cheaply check the output. “Needs setup” means access or reliable records are missing. “Keep human-led” covers commitments, sensitive judgments and decisions where an error would be costly. These are your decisions, not the assistant’s permission to act.

Copy this assignment:

4. Run one trial. Have the assistant prepare a meeting brief from that same folder. Check every deadline and named commitment against its source, and note how long checking took. Keep the workflow only if the reviewed result is accurate and the total effort beats doing it yourself. This is a proposed exercise, not a tested time-saving claim.

If you manage a team, compare maps to find shared access problems before buying more software. For today, choose one folder you are authorized to use and one result you could verify yourself.

TOOL SPOTLIGHT

Plaud Pro costs $17.99 a month; Work mode uses separate credits

For: People who already use Plaud for regular meetings or interviews and want to evaluate the follow-through features described above.

Cost: Plaud’s current U.S. plan page lists a free Starter plan with 300 transcription minutes a month, included with its devices. Pro is $17.99 billed monthly for 1,200 minutes; annual billing is advertised at $8.33 a month. Hardware is a separate purchase.

The catch: Transcription minutes and agent work are different meters. Work mode consumes Plaud Credits; the one-time preview grant is not unlimited ongoing use. Exact top-up pricing was not verified. The plan page labels several connectors, including Slack, Notion and Linear, as previews.

Skip it if: Occasional written notes already do the job, or workplace policy does not permit recording and uploading these conversations. Existing customers should test one authorized, low-sensitivity recording before upgrading.

Testing status: Product overview based on published documentation, not hands-on testing.

BEFORE YOU GO

Choose one recurring task for AI to prepare and you to review

What is one recurring task you would trust AI to prepare, as long as you could review it before anything happened? That is a useful place to start this week.

—Aric