Newsletter

Obizworks AI Brief #5

By Obizworks Editorial · 2026-11-20 · 6 min read
🛈 Governed AI, human-reviewed. Drafted by Obizworks' governed AI and reviewed by a human before publication.

Welcome to the Obizworks AI Brief: one short email a week on what changed in AI, what it means for a small or mid-sized business, and a few things you can try. This issue covers 9 to 15 November.


The big picture

AI agents started knocking on the front door this week. ChatGPT began letting businesses claim a verified profile and run their own agent inside it, while a major US airline moved to keep outside agents off its website. At the same time, AI tools moved further into everyday office work. The lesson for a growing business: decide what agents may do on your behalf and on your site, and match the most powerful (and most expensive) AI to the jobs that are worth it.


What happened

1. Businesses can now claim a profile in ChatGPT. OpenAI launched "Businesses in ChatGPT" in beta. Companies can claim a verified profile through a new Business Hub, retailers can add product feeds, and some businesses can set up a Business Agent that talks to customers inside ChatGPT. Agents start with eligible US ecommerce and online-service businesses, must say they are AI, and businesses see only summary figures about the conversations. Why it matters: customers increasingly ask an AI before they visit your website. Check what ChatGPT says about your business today, and claim your profile when it becomes available to you.

2. United Airlines moves to keep outside AI agents off its website. United said it will restrict unauthorised third-party personal AI agents from its website and app, including agents working through online travel sites. Its reasons include shared passwords, purchases or points redemptions customers did not intend, and check-in steps that rules say a person must confirm. Why it matters: every business with online booking, ordering or accounts will face the same question. Decide which steps an outside agent may take, and which a person must always confirm.

3. Claude adds spreadsheets, whiteboards and dashboards. Anthropic put Claude Sheets and Claude Whiteboards into beta and made Claude Data, which builds dashboards from connected business systems, generally available. They are on paid plans. Sheets builds spreadsheets with formulas and charts from a conversation and can export to Excel. Why it matters: AI assistants are becoming places where office work gets done, not just where questions get answered. Before staff connect business data, decide which systems they may connect and who can see the results.

4. A more capable model, at a premium price. Anthropic released Claude Fable 5.2, priced at $8 per million input tokens and $40 per million output tokens, about 20% less than the previous version. Anthropic still advises starting with its cheaper Opus 5.5 model for most work and moving up only when a task needs it. Why it matters: the best model is not the right model for every job. Use the expensive one where the result is worth a lot, and route routine work to cheaper models.

5. IBM and Microsoft launch a catalogue of ready-made agents. IBM and Microsoft launched "Imagine Enterprise", a platform with more than 6,000 prebuilt agents and assets from IBM Consulting, starting with financial services, insurance and retail. No pricing was published. Why it matters: large companies are buying agents off the shelf. A smaller business can borrow the idea: start from a proven pattern for one workflow instead of building from scratch.

6. A coding assistant setting leaked passwords and keys. Security firm Apiiro reported that a "don't ask again" approval option in Claude Code can save commands, including any passwords or keys inside them, to a settings file. Developers then committed those files to public GitHub repositories, exposing thousands of secrets. Why it matters: AI tools copy whatever you give them into places you may not check. Keep passwords and keys out of commands and code, and scan your repositories for exposed secrets.


From the podcasts

Nate B Jones, AI News & Strategy Daily, on Claude Fable 5.2 (12 Nov). After a month of failed attempts with other models, Nate says Fable 5.2 read his earlier work, worked out what he wanted and built it. It even suggested starting a fresh session with a handoff file when the conversation got long. His bigger point is cost: many businesses find their current model already handles most tasks, which he calls "task-level saturation". He describes a K-shaped split: people who find the hard, valuable problems pull ahead, while others see little change. Our take: before paying for the strongest model, name the job where getting it right is worth many times the bill. For everything else, a cheaper model is fine.

Nate B Jones, on chips, China and AI speeding up AI (10 Nov). Nate covered the questions around Nvidia chip sales to China, and then turned to METR, a research group that spent 11 weeks with Anthropic's internal data. METR estimated that Anthropic's models are now improving roughly twice as fast as before, with a plausible range of 1.3 to 3 times, mostly because AI is helping with AI research. METR noted that Anthropic redacted much of the evidence from the public version. Nate also pointed to the Binance theft, where attackers reached wallet engineers' computers through a fake job interview and then used an internal AI tool. Our take: plan for AI to keep changing faster than your contracts. And treat every internal AI tool as a door: give it only the access it needs, and log what it does.

Moonshots with Peter Diamandis, #311 (13 Nov). The panel discussed Demis Hassabis of Google DeepMind on "recursive self-improvement". Today, he said, AI makes human engineers several times more productive while people still review the work. The riskier next step would be closing the loop, with AI improving AI without human review. The panel also discussed Anthropic's new office tools and the pressure they put on established software makers. Our take: the leaders building AI are arguing for keeping a person in the loop at the frontier. The same rule works in a small business: AI drafts, a person approves.


Three things to try this week

1. Ask ChatGPT about your business. Ask what it knows about your company and your competitors. Note anything wrong or missing, and check whether you can claim a business profile yet.

2. Write a one-page agent policy. List what an outside AI agent may do on your website or booking system, and which steps (payments, bookings, account changes) a person must always confirm.

3. Hunt for exposed keys. Ask whoever manages your code or website to scan your repositories for passwords and keys, and to replace any they find.


On the Obizworks blog

Sources