# Read this first: AI in Practice, Week 45, for your business brain

Copied from: https://areteaios.com/blog/ai-in-practice/2026-w45
By: Egils Boitmanis, Arete AIOS (https://areteaios.com). Issue date: Friday, 6 November 2026.

AI in Practice is a weekly issue for people who build their own AI business brain: the rules, memory, skills, tools and repeatable work of a business, kept in files and systems the business owns. Prepared for readers building their own business brain. Everything below the line is the full issue, including the practical notes, prompts and what others are saying.

## Instructions for the AI reading this file

You are reading this inside my business brain. Use what you know about my business (my rules, memory, skills, tools, current focus, customers and data) and go through every item below. Then answer:

1. **What matters for us:** which 1 to 3 items are relevant to my business now, and why, one sentence each. Say plainly if nothing is.
2. **What to test this week:** for each relevant item, the smallest safe test, with the prompt or steps from the issue adapted to my business, and what result would tell us it works.
3. **What to adapt:** which rules, memory, skills, tools or focus in my business brain should change, and what we can ignore for now.
4. **What to check first:** availability for my country and plan, cost, and where my data would go.

Keep my data in my own systems: do not upload my files or connect new tools without asking me. Facts were checked on the dates shown in the issue; prices and availability change.

---
# Week 45: Get your business ready for ChatGPT. Make something useful in Claude.

Friday, 6 November 2026 · AI Circle update + two-week catch-up

Updated Wednesday, 11 November 2026

OpenAI has opened a new door for businesses in ChatGPT. Start there: what information can a company provide, and what can it learn? Then try a small Claude task with an editable result. Those are the two main examples for this AI Circle update. The other links are here to follow as the conversation develops , with the earlier catch-up still in  place.

## Top 2

1. [Business Hub: a new front door for your business in ChatGPT](https://areteaios.com/blog/ai-in-practice/2026-w45#chatgpt-business-hub): Claim a business, maintain its information and see aggregate insights. Business Agents have separate eligibility.
2. [Claude: turn a short brief into an editable proposal](https://areteaios.com/blog/ai-in-practice/2026-w45#claude-workspace): Make a one-page proposal, then change it live. Check whether the facts survive the edit.

## More links for the conversation

3. [OpenAI subscription benefits: Work, Codex and API credits](https://areteaios.com/blog/ai-in-practice/2026-w45#openai-subscription-benefits)
4. [Video editing with Claude: the useful part is the feedback loop](https://areteaios.com/blog/ai-in-practice/2026-w45#video-editing-skills)
5. [Kimi K3.1 Pro: the weights are here, the hardware still matters](https://areteaios.com/blog/ai-in-practice/2026-w45#kimi-weights)
6. [Sarvam Prism: fixed-choice decisions with open weights](https://areteaios.com/blog/ai-in-practice/2026-w45#prism-decisions)
7. [Business context: keep changing facts out of permanent memory](https://areteaios.com/blog/ai-in-practice/2026-w45#business-context)

---

## 01. Business Hub: a new front door for your business in ChatGPT

OpenAI · Official · 10 Nov 2026

OpenAI introduced Businesses in ChatGPT in beta on 10 November. Business Hub lets eligible companies claim a profile, maintain business information and provide product feeds. It also offers aggregate insights such as mentions, website clicks and discovery topics, subject to activity thresholds.

Claiming a business does not guarantee a recommendation. Business Agents are a separate feature, initially for eligible US ecommerce and online service businesses. A profile, an agent and advertising have different requirements. That distinction matters before you plan what to show.

![Business Hub overview: company information and aggregate insights, with Business Agents shown as a separate feature](https://areteaios.com/ai-in-practice/2026-w45/business-hub.svg)

Source: [OpenAI: Introducing Businesses in ChatGPT](https://openai.com/index/introducing-businesses-chatgpt/). Arete AIOS explanatory diagram based on OpenAI’s 10 November announcement and setup guide. It is not an account screenshot.

### Practical usage

**How this affects business brain builders:** Use your business brain to prepare consistent public facts: what you offer, who it is for, your website and current policies. Check them before submitting.

**Prompt: Prepare the business facts before opening the Hub**

```text
Using only the public business information I provide, prepare a short company description, our main offers, intended customers, website and relevant policies. Separate confirmed facts from missing information. Identify contradictions. Do not invent prices or claims. Return a concise draft for me to review before entering it in Business Hub.
```

Try: [Open ChatGPT Business Hub](https://chatgpt.com/business-hub)

Access: Checked 11 Nov 2026: availability depends on business location, type and supported verification. Claim and Persona approval are required for management features. This account’s eligibility and metrics have not been checked.

Status: Not tested yet

1. Open the Hub in your own account and inspect the business search and options available to you.
2. Compare the information shown with your current public website. Identify one correction worth making.
3. If insights are available, examine a metric and its definition. A blank metric can reflect a threshold, rather than zero activity.

**Can I see customers’ conversations?** No. Insights are aggregate and subject to minimum activity thresholds.

**When does a new profile become public?** The setup guide says an approved new business profile publishes automatically after the required verification. Inspect the flow first; submitting a claim is a separate decision.

### What others are saying

- The guide explains eligibility, verification, automatic publication of approved new profiles and the limits of aggregate insights. Independent practitioner results are not established in this update. [OpenAI Help Center, 11 Nov 2026](https://help.openai.com/en/articles/20001635-setting-up-your-business-on-chatgpt)


---

## 02. Claude: turn a short brief into an editable proposal

Anthropic · Recap · as of 6 Nov 2026

Claude’s unified experience brings quick questions and longer Cowork tasks into one conversation. You can ask for a report, a spreadsheet or a presentation without choosing a separate mode first. The rollout is gradual, so two people on the same plan may still see different interfaces.

The new practitioner videos make this worth revisiting. They show artifacts and ways to organize several jobs at once. Some capabilities were announced back in September. This is a practical catch-up on the working environment, not the launch of a model called Claude 2.0.

![Futurepedia video thumbnail for its 4 November 2026 overview of the Claude desktop experience](https://areteaios.com/ai-in-practice/2026-w45/claude-overview.jpg)

Source: [Anthropic: Claude Cowork and chat are one Claude](https://support.claude.com/en/articles/16761823-claude-cowork-and-chat-are-one-claude). Futurepedia’s 4 November walkthrough. The thumbnail identifies the source video; it is not a verified screenshot of a particular account’s feature access.

### Practical usage

**How this affects business brain builders:** A more capable workspace still needs a clear handover: the outcome, the business context, the files it may use and the decisions that remain yours.

#### The new workspace, explained by a practitioner

![Futurepedia source video thumbnail](https://areteaios.com/ai-in-practice/2026-w45/claude-overview.jpg)

Futurepedia, 4 Nov. Relevant sections: 01:40–03:25 for artifacts and 09:00–12:15 for organizing work. This issue reviews the transcript; it does not claim visual validation of every feature.

Source: [Futurepedia: Claude overview](https://www.youtube.com/watch?v=tM0dwXhOv2o)

[Watch video](https://www.youtube.com/watch?v=tM0dwXhOv2o)

**Prompt: A small live demo in Claude Docs**

```text
Create an editable one-page proposal in Claude Docs from this fictional training brief. Bright North offers a 90-minute introductory AI workshop for a team of up to 12 people. The workshop includes two practical exercises and a short follow-up checklist. The client is a small accounting firm. No date, price or delivery location has been agreed. Explain the outcome, agenda, included deliverables and decisions still needed. Keep it under 350 words. Use plain English and do not invent missing commercial terms. After creating the document, wait for my feedback.
```

Try: [Check artifact availability by plan](https://support.claude.com/en/articles/17153992-what-are-artifacts-and-how-do-i-use-them)

Access: Checked 11 Nov 2026: Anthropic lists Docs, Slides and Design on all plans; Data is paid-plan access, and Motion is a Team/Enterprise beta. Artifacts require the file-creation capability. Availability in this editor’s account has not been tested.

Status: Not tested yet

1. Open Claude Docs and use the fictional brief above, or a public brief you can share.
2. Ask: shorten this by one third, remove jargon and make the decisions still needed easier to find. Inspect the edited document.
3. Check that the 90-minute duration, 12-person limit and missing price/date survived. Export only after reviewing the result.

**Can one Claude session see every other session?** The desktop Code-tab management surface covers its local, SSH and WSL sessions. It excludes cloud, terminal CLI and VS Code sessions; by default it sees the 20 most recent eligible sessions.

**Are all the features in the video available to everybody?** No. Plan, beta status and staged rollout matter. Futurepedia’s expectation of wider Motion access is not a confirmed entitlement.

### What others are saying

- The walkthrough focuses on artifacts, context visibility, session organization and remote work. Its Claude 2.0 title describes the author’s impression of the app, not a model release. [Futurepedia, 4 Nov 2026](https://www.youtube.com/watch?v=tM0dwXhOv2o)
- Wolfe uses a project and reusable skills to turn a general assistant into a specialized video workflow. His setup also exposes the practical friction around tool access and permissions. [Matt Wolfe, 4 Nov 2026](https://www.youtube.com/watch?v=ep4psgAFOqc)


---

## 03. OpenAI subscription benefits: Work, Codex and API credits

OpenAI · Official · 28 Oct 2026

Eligible ChatGPT Pro and Business subscriptions now include an API credit benefit. Pro $100, $200 and $500 tiers receive the matching monthly dollar amount after activation. For someone building a small agent or business tool, that can make an experiment possible without a separate initial API purchase.

There are two separate meters here. Work and Codex share subscription usage. API credits pay for eligible API Platform usage and do not raise your ChatGPT or signed-in Codex limits. Codex usage through API keys is excluded from this benefit.

![OpenAI documented monthly API credit allowances by subscription, checked 6 November 2026](https://areteaios.com/ai-in-practice/2026-w45/api-credit-benefit.svg)

Source: [OpenAI: API credits for ChatGPT subscribers](https://developers.openai.com/api/docs/guides/chatgpt-api-credits). Arete AIOS summary of OpenAI’s published allowances. This is not a screenshot of an account or its available balance.

### Practical usage

**How this affects business brain builders:** Start with a small, capped task your business brain already understands. Compare the saved time with the setup and verification work.

**Prompt: Plan an experiment (any business brain; no API call)**

```text
Find one small repetitive task in my business that could benefit from an API workflow. Define the input, desired output, a pass/fail check and a strict cost ceiling. List the permissions and data needed. Do not create an API key, activate a benefit, connect an account, spend money or run the experiment.
```

Try: [Read the benefit and eligibility rules](https://developers.openai.com/api/docs/guides/chatgpt-api-credits)

Access: Checked 6 Nov 2026: an eligible paid Pro or Business subscription at least seven days old, matching account email and ownership of the linked API organization are required. Account eligibility and regional availability have not been verified here.

Status: Not tested yet

1. Check your plan, remaining subscription usage and benefit eligibility in your own account.
2. Pick one repetitve task with a small input and a result you can check.
3. Set the spending boundary before authorizing an implementation.

| Meter | What it covers |
|---|---|
| Subscription usage | Work and signed-in Codex share an allowance. |
| API benefit | Eligible API Platform usage in the linked organization. |
| This review | Documentation checked; no benefit activated and no account balance verified. |

**Does this mean another doubling of Work or Codex limits?** No additional doubling is established by this API benefit. The current pricing page points you to your own usage dashboard for your actual subscription limits.

**Can unused credits wait until next month?** They expire with the billing cycle. Check the first grant and its expiry date after linking.

### What others are saying

- OpenAI separates API token pricing from included subscription usage. Its pricing guide directs readers to their own usage dashboard for actual limits. [OpenAI documentation, 6 Nov 2026](https://learn.chatgpt.com/docs/pricing)


---

## 04. Video editing with Claude: the useful part is the feedback loop

Matt Wolfe · Post · 4 Nov 2026

Matt Wolfe’s new walkthrough starts with his own talking-head recording and supporting material. Claude prepares a script and shot list, cuts the recording and assembles a short video. His stack uses VidIQ, Hyperframes, FFmpeg and ElevenLabs. VidIQ sponsors the video.

The first attempt does not simply work. There are transcription and website-access blocks, missed mistakes, gaps and unwanted voice processing. Wolfe gives specific feedback, gets another version and asks for the improvements to be saved in the skill. He still prefers more hands-on editing for long videos.

![Matt Wolfe source video thumbnail: How I Get Claude To Edit My Videos](https://areteaios.com/ai-in-practice/2026-w45/matt-editing.jpg)

Source: [Matt Wolfe: the full workflow](https://www.youtube.com/watch?v=ep4psgAFOqc). The original tutorial, published 4 November. Its reported results belong to Wolfe; this issue has not run the editing workflow or inspected the finished video visually.

### Practical usage

**How this affects business brain builders:** Teach the system how to make one useful output, review its mistakes and preserve the correction. That is a repeatable business process.

#### Watch the first attempt and its correction

![Source video thumbnail for the editing walkthrough](https://areteaios.com/ai-in-practice/2026-w45/matt-editing.jpg)

At about 20:00–25:05, Wolfe describes the first edit’s problems, gives feedback and presents a revised version. Review those frames and the audio before judging quality.

Source: [Matt Wolfe: first edit and revision](https://www.youtube.com/watch?v=ep4psgAFOqc&t=1200)

[Watch video](https://www.youtube.com/watch?v=ep4psgAFOqc)

**Prompt: Plan one short-video editing test (existing editing environment)**

```text
Help me plan an editing test from a 60–90 second recording I will choose. First produce an edit decision list: mistakes, repetitions, pauses, captions and optional supporting visuals. Use transcript timing and the audio waveform to avoid cutting words. Keep the original untouched. Use only existing authorized tools. Ask before paid services or external uploads. Return the plan and quality checks before rendering.
```

Try: [Watch the original editing example](https://www.youtube.com/watch?v=ep4psgAFOqc&t=1200)

Access: The source uses paid and connected services. Access, price and availability vary by account and region. This is a proposed test using an existing authorized editing environment, not a claim that every step is included in a Claude subscription.

Status: Not tested yet

1. Choose a short recording with a known beginning and ending.
2. Review the proposed cuts, then compare the first version with the original sound.
3. Save only the corrections you want repeated in the next edit.

**What should count as a successful test?** No clipped words or changed meaning, intelligible natural voice, correct captions and a visible list of edits. A faster render alone is not enough.

**Does the tutorial prove fully automatic long-form editing?** No. The demonstrated focus is short talking-head content, and the author explicitly keeps an editor in his long-form process.

### What others are saying

- The tutorial includes the blocks and the imperfect first edit. Wolfe reports that a later revision improved the cuts and sound after specific feedback. [Matt Wolfe, 4 Nov 2026](https://www.youtube.com/watch?v=ep4psgAFOqc&t=1260)
- Futurepedia says it uses code-generated motion graphics in its own videos, through an editing platform it built. That is a different workflow from generating an entire filmed video. [Futurepedia, 4 Nov 2026](https://www.youtube.com/watch?v=tM0dwXhOv2o&t=180)
- Berman relays positive video-editing feedback about GPT-6.1 Terra from two people on his team. It is a team report, not a controlled comparison with Wolfe’s setup. [Matthew Berman, 5 Nov 2026](https://www.youtube.com/watch?v=-M62vWrZh54&t=1094)


---

## 05. Kimi K3.1 Pro: the weights are here, the hardware still matters

Moonshot AI · Recap · as of 6 Nov 2026

In the previous review, full Kimi K3.1 Pro weights were still a promise. The official repository is now public and lists 96 weight files totaling about 1.56 TB. The model uses a custom licence, so check its terms before commercial deployment.

Moonshot’s research compares the model with proprietary systems across coding and knowledge work. Those are vendor results. For a small business, the useful decision is whether a hosted provider or controlled infrastructure fits a specific task; a model this size is not a normal laptop install.

![Moonshot’s original BenchCAD comparison, with and without tools, on a 1,000-file subset](https://areteaios.com/ai-in-practice/2026-w45/kimi-benchcad.webp)

Source: [Moonshot: original research and comparisons](https://www.kimi.com/en/blog/kimi-k3-1-pro). Moonshot’s original BenchCAD chart, on a 1,000-file subset using Voxel IoU. This is a specific CAD-reconstruction test, not overall business performance.

### Practical usage

**How this affects business brain builders:** Use one fixed business task to compare output quality, total running cost and control over the data. A model’s headline rank cannot make that choice for you.

#### The broader vendor comparison

![Moonshot’s full Kimi K3.1 Pro benchmark overview](https://areteaios.com/ai-in-practice/2026-w45/kimi-benchmarks.webp)

The full release compares several coding, agent and knowledge-work evaluations. Check the benchmark-specific notes and reasoning settings before comparing scores.

Source: [Moonshot: full release and methods](https://www.kimi.com/en/blog/kimi-k3-1-pro)

**Prompt: Compare deployment options (planning only)**

```text
For one recurring task in my business, compare a hosted frontier model, a hosted open-weight model and infrastructure I control. Use the same input and pass criteria. Include data handling, licence, memory, maintenance and cost per completed task. Mark unknowns. Do not download weights, provision a server, send business data or start paid calls.
```

Try: [Inspect the official files and licence](https://huggingface.co/moonshotai/Kimi-K3.1-Pro/tree/main)

Access: Checked 6 Nov 2026: the official repository is public and ungated, with a custom licence and roughly 1.56 TB of weight files. No model was downloaded or run in this review. Hosted access has separate terms.

Status: Not tested yet

1. Pick a task with an answer you can independently check.
2. Check licence and data-handling terms for each deployment option.
3. Compare the complete cost of a correct result, including setup and maintenance.

**What changed since the previous issue’s research?** The weights can now be found in the public official repository. The earlier announcement date has not become a new model launch date.

**What does the chart measure?** Moonshot combines several coding, agent and knowledge-work evaluations. The full release article gives settings and benchmark-specific qualifications; there is no single score for your business.

### What others are saying

- Berman highlights the public weights, then immediately points out their size. His suggestion that smaller quantized versions may arrive is an expectation, not an already available product. [Matthew Berman, 5 Nov 2026](https://www.youtube.com/watch?v=-M62vWrZh54&t=37)
- Wolfe separates the value of controlled enterprise infrastructure from the dream of running a frontier model on a home computer. This large model does not make those the same thing. [Matt Wolfe, 30 Oct 2026](https://www.youtube.com/watch?v=eDiQNFa-zYg&t=819)


---

## 06. Sarvam Prism: fixed-choice decisions with open weights

Sarvam AI · Open source · 3 Nov 2026

Sarvam released Prism, a decision model built on Sarvam-105B. Give it context and questions with fixed answer options. It returns probabilities for those options rather than generating an open-ended text answer. The published model card lists an Apache 2.0 licence.

That suggests a different design for repeated classification work: define the possible actions and a human-review threshold. Sarvam’s comparison includes Jev, but its benchmark and latency claims are vendor measurements. They do not establish performance on your inbox, language or business rules.

![Selected rows from Sarvam’s published model comparison, with latency qualifications](https://areteaios.com/ai-in-practice/2026-w45/prism-comparison.svg)

Source: [Sarvam: release, method and comparisons](https://www.sarvam.ai/blogs/sarvam-prism). Selected rows from Sarvam’s original published table, typeset by Arete AIOS. Vendor evaluation; API-only reported latency and locally measured latency have different conditions.

### Practical usage

**How this affects business brain builders:** Keep a cheap repeated decision separate from the more expensive job of writing, explaining or acting on it. Send uncertainty to a person.

**Prompt: Design a fixed-choice test (any business brain)**

```text
Choose one repetitive classification task from my business. Define a short list of answer options, including unknown or needs review. Create 20 clearly synthetic examples and an answer key. Propose an acceptance threshold and how to measure errors. Do not process real messages, connect an API, send anything or run a paid model.
```

Try: [Read the model card and requirements](https://huggingface.co/sarvamai/prism)

Access: Checked 6 Nov 2026: open weights are published under Apache 2.0. The model card also links to a hosted API. Local inference needs substantial GPU resources; hosted prices, account access and regional terms are not verified here.

Status: Not tested yet

1. Define mutually understandable categories and a needs-review option.
2. Prepare synthetic examples with a separate answer key.
3. Measure wrong decisions and escalation behavior before allowing any real action.

**Why not ask a normal chat model?** A fixed-choice task can be a narrower component in a larger process. Whether it is better or cheaper depends on your test, volume and deployment.

**Are the comparison numbers independent results?** No. Sarvam ran and reported the comparison. Its selected indices and hardware conditions matter; no local reproduction was done for this issue.

### What others are saying

- Berman connects Prism with the decision-model approach used by Jev. He also points to hardware and model-size requirements, rather than presenting it as a free ordinary-laptop replacement. [Matthew Berman, 5 Nov 2026](https://www.youtube.com/watch?v=-M62vWrZh54&t=434)


---

## 07. Business context: keep changing facts out of permanent memory

AI in Practice · Recap · as of 6 Nov 2026

Mansel Scheffel’s new explanation starts with a familiar problem: a powerful model has not been briefed on the business. He separates repeatable methods, knowledge, live data and state, and the small set of things an assistant should remember across work.

The practical distinction matters. A customer’s unpaid invoice is changing state. A rule for checking overdue invoices is a reusable method. Mixing them in permanent memory makes yesterday’s truth look like a standing instruction. Keep the source of a fact clear and review what has become stale.

![Mansel Scheffel video thumbnail for the business context explanation](https://areteaios.com/ai-in-practice/2026-w45/context.jpg)

Source: [Mansel Scheffel: the context and memory explanation](https://www.youtube.com/watch?v=uo27pKVKnEA). Source video published 4 November. This issue reviews the opening and memory-maintenance chapters, not the whole video or its visual demonstrations.

### Practical usage

**How this affects business brain builders:** Keep the business context you want to carry between tools understandable, current and attached to its actual source.

**Prompt: Audit the context already available (any business brain)**

```text
Review only the business context already available to you. Separate stable knowledge, reusable methods, changing records and conversational preferences. Flag duplicated or stale statements, with a source and reason. Propose what to keep, move into a skill or retire. Do not change files, delete memory or read additional accounts; return the proposed changes for review.
```

Try: [Watch the relevant memory chapter](https://www.youtube.com/watch?v=uo27pKVKnEA&t=1740)

Access: The audit prompt can be used in an existing AI workspace with access to the chosen context. Tool-specific file and memory controls differ. No new database, connector or automatic schedule is required by this exercise.

Status: Not tested yet

1. Select a small set of business notes or instructions you already use.
2. Separate durable guidance from facts that change with each transaction.
3. Review the proposed corrections before applying them to the source.

**Was this a product release?** No. It is a practical explanation selected for the business-brain question behind the other stories.

**What part of the source was reviewed?** The caption text for 00:00–04:59 and 29:00–33:59. The full caption file is available, but the rest was not reviewed for this issue.

### What others are saying

- Scheffel’s maintenance loop is to observe, review and decide what to retain, turn into a reusable method or retire. He warns about changing customer state lingering in permanent memory. [Mansel Scheffel, 4 Nov 2026](https://www.youtube.com/watch?v=uo27pKVKnEA&t=1740)
- The editing workflow gives a concrete companion example: a correction becomes useful across future jobs when it is preserved in the editing skill. [Matt Wolfe, 4 Nov 2026](https://www.youtube.com/watch?v=ep4psgAFOqc&t=1440)

---

Pick one task you already do. Give it a clear finish line, use the tools you have and check the result against the source. The useful improvement is the one you can repeat. Questions or a result to share: egils@areteaios.com.
