Claude's latest update is less about one model score than about where work happens. In Riley Brown's weekly walkthrough, a conversation becomes an editable whiteboard, a working spreadsheet and a live data view; a meeting can create those artifacts while people talk. He then moves through Fable 5.2, Google's Gemini 4 Carbon and OpenAI's expanding Dots. The important distinction is between a feature Riley used, a provider's stated rollout, and an interpretation of what it means.
Claude Whiteboards and Sheets are paid-plan betas; Data is generally available with a connector. Meetings is a limited Team/Enterprise beta for Google Meet. Fable 5.2 is broadly available, while Mythos 5.2 requires verification. Dots can now use multiple local computers, but Wallet starts with eligible US Pro users. United's restriction concerns unauthorized third-party agents, and its stated reason is security and compliance.
Watch Riley Brown's Full Update
Riley's screen recording is valuable because he actually revises a whiteboard and builds a sheet, then shows his two-computer Dot setup. His model comparison and comments about companies' motives are his impressions, not controlled tests or official policy explanations.
Claude's New Workspace: Three Different Jobs
| Tool | What it does | What to know before using it |
|---|---|---|
| Whiteboards | Turns a discussion into a board that people and Claude can rearrange, annotate and convert to other artifacts. | Beta on paid plans. Good for mapping a workflow or assigning owners; do not mistake a generated diagram for a verified process. |
| Sheets | Builds and edits tabbed spreadsheets with formulas, formatting and drop-downs. | Beta and currently simpler than a full spreadsheet suite. Export to .xlsx or CSV when advanced Excel features are needed. |
| Data | Queries connected sources and shares dashboards with current numbers, provenance and approved views. | Generally available on paid plans, but requires a data connector. Certified queries and live sharing are still beta features. |
Riley's own example is concrete: he asks Claude to review many past video-editing threads, turn them into a visual workflow, and then make a categorized resource sheet with charts. He switches from Fable 5.2 Extra to Opus 5.5 at medium effort mid-demo because the job does not justify the larger model. That is a useful cost lesson, not evidence that the models perform identically. Watch the artifact walkthrough. Anthropic's launch guide confirms the feature and rollout details. Team admins have these enabled by default; Enterprise admins may need to enable Whiteboards and Sheets. Data depends on the connector and its permissions.
The practical pattern is: start with messy context, ask for a board to expose the structure, then build a sheet for inventory and a connected Data view only when the numbers need to stay current. Review formulas, source mappings and sharing permissions before a dashboard becomes a decision system.
Claude Meetings: An Active Participant, with Consent
At 05:09, Riley watches Anthropic's meeting demo: Claude joins as a visible participant, follows the discussion, answers questions and starts documents or whiteboards during the call. This goes beyond after-the-fact transcription, but it is not an unrestricted teammate. Anthropic says connected-tool actions require permission, and the product is rolling out in beta to Team and Enterprise, currently for Google Meet. The meeting page keeps a transcript, summary and screen descriptions, not the audio recording. Participants should consent to transcription; outside-organization calls have additional default restrictions. Some high-security configurations are excluded from the beta.
A sensible first use is a low-risk internal planning call: have Claude maintain an action list and draft a board, then let a person check owners, deadlines and any tool action before it becomes authoritative.
Fable and Mythos Are Real; Carbon Is a Separate Test
Anthropic launched Fable 5.2 and Mythos 5.2 on 9 November. They share an underlying model, but Fable has stronger cyber and biology safeguards for broad release, while Mythos access is limited to organizations accepted into Anthropic's Cyber or Life Sciences Verification Programs. Fable 5.2 is available to eligible paid subscribers and developers; Anthropic lists API pricing at $8 per million input tokens and $40 per million output tokens. Anthropic itself recommends starting most coding and knowledge work with Opus 5.5 and reserving Fable for ambitious, open-ended technical work.
Riley reports that Fable brings more refinement to his 3D and video work but burns through his Fable-specific limits. His preferred workflow is to use Fable for the hard plan or urgent, quality-sensitive pass, then let Opus execute more routine work. This is his working method, not a guarantee that every two-model handoff saves money or preserves quality. Review the plan and final output separately.
Google's Gemini 4 Carbon is also an official release. Riley says he spent about two and a half hours testing it and still preferred Opus for his own use cases. That is a first impression, not a broad benchmark verdict. For a purchase or migration decision, compare both models on your actual files, prompts, effort levels, latency and total reviewed cost. Watch his Carbon segment.
GrokBot: More Ways to Reach Work, More Permissions to Review
Riley's short GrokBot update highlights Mac app access, Slack improvements and Tailscale connectivity to private tools. He presents these as changes he is tracking, not as a full setup or security audit. Verify the exact app permissions and rollout in your GrokBot account before using them for company systems. For a private MCP server, prefer narrowly scoped access, a dedicated identity and logs that show what the agent actually changed. The practical benefit is reach; the corresponding risk is a wider action surface.
One Dot, More Than One Computer
Riley demonstrates a Dot connected to his MacBook Pro and Mac Mini, asking it to inspect the Downloads folder on each. OpenAI's Dots announcement confirms that multiple local computers can connect to the same Dot and that users can choose the machine for a task. This is particularly useful when a dedicated test or render computer should work while the main laptop stays free. Give each machine only the file and app access the task needs; inspect the Dot's activity before trusting a result from either machine. Watch the two-computer demo.
Riley also perceives faster, less jargon-heavy responses in voice. That is subjective. OpenAI's Dots announcement says GPT-6.1 Terra joined the model family powering Dots on 3 November, but a Dot can draw on different models for different parts of its work. It is inaccurate to say every Dot response is exclusively Terra. Watch Riley's speed discussion.
ChatGPT Wallet: Approval Is a Step, Not an Order Confirmation
OpenAI's 13 November launch lets an eligible Dot prepare a purchase and use a tokenized payment credential after the user reviews the merchant, items, total and card and passes a security check. The agent and merchant do not receive the user's actual card details. Riley says explicitly that he had not tested the feature yet; the shopping sequence in his episode is OpenAI's demo. Watch that segment.
The rollout is initially for US ChatGPT Pro users aged 18 or older with an eligible US-issued card and US billing address, through Dots. It is not a general ChatGPT or Portugal-wide wallet launch. OpenAI's release notes warn that approving a purchase does not itself confirm an order; check the merchant's confirmation. Businesses also need to consider whether their checkout and fraud systems accept legitimate agent purchases.
Why United Is Restricting Unauthorized Agents
Riley reads United's change as part of a broader conflict between shopping agents and companies' incentives, then speculates that price discrimination may be a motive. That is his inference, not a documented finding about United. United's 11 November statement gives a different, explicit rationale: credential exposure, unintended purchases or account and mileage changes, and potential aviation compliance issues such as human acknowledgment at check-in. It says the restrictions target unauthorized third-party personal AI agents on United's platforms and through agencies and aggregators. Watch the discussion.
The tension is real: OpenAI is building an authorized payment path, while a travel provider can still limit unauthorized automation on its own systems. The precise policy and permitted integrations matter more than a blanket claim that agents can or cannot book travel.
Recommended Reading: What Model Weights Can Reveal
Riley closes with Will DePue's checkpoint-forensics report. DePue says numerical patterns in six published models from DeepSeek, Moonshot AI, MiniMax and Ant Group match NVIDIA Blackwell-class hardware, and he describes checks against known models and physical chips. The method looks at how model weights appear to have been initialized. His own report says that this supports, but does not prove, Blackwell use throughout training. It does not establish where chips were located, how access was obtained or whether export controls were violated. Watch Riley's reading recommendation.
That distinction matters because a forensic fingerprint is a research result, not a complete supply-chain or legal investigation. Read the report's limits section before repeating its headline. The original X post was supplied only as a truncated URL; the full report is the stable, verifiable source linked here.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | Introduction |
| 00:36 | Claude Sheets, Whiteboards and Data |
| 05:09 | Claude Meetings |
| 06:56 | Claude Fable 5.2 and Mythos 5.2 |
| 10:49 | Gemini 4 Carbon |
| 12:32 | GrokBot updates |
| 13:24 | Dots connects multiple computers |
| 15:20 | ChatGPT Wallet |
| 17:08 | Faster Dots and GPT-6.1 Terra |
| 17:59 | United restricts AI agents |
| 20:08 | Recommended reading: checkpoint forensics |
Sources and Reporting Limits
- Episode: Riley Brown's video and the supplied transcript. His tests and opinions are attributed to him, not presented as independently replicated measurements.
- Product documentation: Anthropic on workspace artifacts, Meetings and models; Google on Carbon; OpenAI on Dots and Wallet.
- Policy and research: United's statement and DePue's full report. For more hands-on agent workflows, see Agent Native's free resources.
Evidence cutoff: 15 November 2026. Rollouts, eligibility and product prices may change. No independent test of Riley's GrokBot configuration, Wallet purchase or model comparison was performed for this article.