Paul J Lipsky's weekly recap is most useful when read as a guide to where the work should happen. Three new models are available, but his larger point is that the everyday model, the surrounding apps, and the speed of capturing a task may matter more than the latest leaderboard winner. The episode also covers a consequential NVIDIA leadership change, a sponsored notebook connector, Claude's new office tools, and AI agents that now live on a watch.
Use a capable everyday model in the ecosystem you already inhabit, and reserve Fable 5.2 or Carbon for genuinely hard work. Claude's Whiteboards, Sheets and Data make more work possible without an export step. Grok Bot 1.0 is moving toward shared team agents. The watch demo suggests a smaller but practical shift: capture and delegate in seconds, without opening a phone. NVIDIA's CEO transition is real; claims of compromised secrets are not established.
Watch Paul's Weekly Recap
NVIDIA: A Confirmed Transition, Not a Confirmed Breach
NVIDIA announced on 12 November 2026 that Jensen Huang resigned as CEO, president and board member, effective immediately. It named long-serving CFO Colette Kress interim CEO and appointed Harvey C. Jones board chair while it searches for a permanent successor. That is a significant leadership change for the company supplying much of the AI industry's compute.
The personal and intelligence allegations that accompany this story should not be compressed into a proven security incident. NVIDIA's board statement said it had opened an independent investigation and had no reason to believe national security or NVIDIA secrets were compromised. A change at the top does not, by itself, establish delays to GPU deliveries, financing, or future AI model releases. NVIDIA says it expects customer commitments and its product roadmap to continue without interruption; the longer-term effects remain to be seen.
Three New Models, Three Different Jobs
Paul describes Claude Fable 5.2 as especially compelling for coding and visual builds, and Gemini 4 Carbon as strong for writing and demanding knowledge work. Those are his hands-on preferences, not universal rankings. Anthropic recommends starting with Opus 5.5 for most everyday coding and knowledge tasks, then moving to Fable for unusually difficult, open-ended technical work. Google positions Carbon similarly for long-horizon reasoning, while saying Argon remains a strong default for many tasks.
Gemini 4 Hydrogen, released 12 November, changes the daily-driver equation. Google positions it for fast coding, tool use and general tasks, with a one-million-token context window and an API rate of $1.50 input / $7.50 output per million tokens. Paid Google AI Pro and Ultra subscribers can use it in Antigravity; in the Gemini app, free users may reach it through automatic model selection. Paul's day-one impression is positive, but it is still a day-one impression.
| If your job is... | Start here | Escalate when... |
|---|---|---|
| Email, routine research, admin, quick coding | Hydrogen, Opus 5.5, or your current daily driver | The output misses important constraints or needs deep verification |
| Large architecture, hard debugging, ambitious visual build | Opus or Argon for a baseline | Fable 5.2 or Carbon earns its higher cost on your exact task |
| High-stakes report or business decision | One model to draft, source documents to check | A second model can challenge the work; a human signs off |
The best subscription is not necessarily the one with the strongest demo. Choose around the files, connectors, apps, privacy requirements and usage limits you actually use. Anthropic's Fable launch offer includes up to half of Pro and standard Team weekly usage through 16 November at 11:59 p.m. PT; after that, Fable returns to usage credits for those plans. Google's Carbon API introductory rate of $2/$10 per million tokens ends 30 November, then rises to $4/$20. These dates matter more than an unsourced blanket claim that either model is "cheap."
The Kortex MCP Demo Is Sponsored
In the episode's sponsored segment, Paul asks Claude a question about a notebook in Google's Gemini Notebook service, then uses Kortex MCP to send the question to the notebook and bring the answer back to Claude. He also demonstrates saving a note and starting an audio overview. That is a useful cross-app pattern: keep a source collection in one product while working from the assistant you prefer.
Kortex is a third-party connector, not an official Google or Anthropic feature. Its plan guide says MCP access requires Pro or Lifetime. Before connecting client or personal notebooks, review the permissions, account boundary and data handling. Keep any action that changes or shares a notebook behind your approval.
Claude's New Office Suite: Boards, Sheets, Data
Anthropic launched Whiteboards and Sheets in beta and moved its Dashboards product into general availability as Claude Data. Whiteboards are for arranging and editing ideas with teammates. Sheets can build tabbed spreadsheets, formulas and charts. Data connects to business sources and supports approved metrics and shared dashboards. All are listed for paid plans, but availability can depend on rollout and workspace settings.
Paul makes a whiteboard from his video notes and a sheet from a fictional plumbing business's records. The strongest use is not that AI can draw a board or a pie chart; it is that people can inspect, rearrange and correct the material in the same workspace. For financial or operational use, confirm the source rows, formulas and definitions before relying on the visual summary. Anthropic also announced fuller interactive connectors so supported partner apps can open beside Claude. Claude Meetings was described as coming soon in the 10 November announcement and has since begun a Team and Enterprise beta. A demo or rollout announcement does not mean every account has the feature today.
Grok Bot 1.0 Is About Shared Work
Grok Bot 1.0's official release focuses on three things: shared team bots, more proactive follow-up and better browser and document work. The changelog adds practical details on meetings, Microsoft Teams and shared connectors. This is potentially more useful than another model benchmark for a team that repeatedly handles the same inbox, meeting or research workflow.
Start with a single narrow bot, a named owner and a reviewable task log. Do not give a shared agent blanket permission to email clients, move money, or change records just because it can operate the browser. The provider says it has improved approvals and efficiency, but your own testing should establish whether those gains hold for your accounts and processes.
The Smaller Interface That Changed Paul's Day
Paul's most personal takeaway is the Muse watch app. He uses dictation to capture a video idea or task while walking, then lets the agent continue after he closes the watch interface. Muse's Apple Watch app appears in Meta's App Store listing; a Wear OS version was also announced. The watch is a convenient input and notification surface, not evidence that the model itself runs locally on the watch.
This is a sensible experiment for people who repeatedly lose small ideas between meetings: create a dedicated capture destination, dictate one task, then verify it landed in the right place. Keep purchasing, outbound messages and sensitive changes approval-gated. Paul also notes that texting an OpenAI Dot from a watch can offer a similar capture path without a native watch app. Whether it improves focus, as it has for him, is personal rather than a measured universal benefit.
The Rapid-Fire Updates, Organized
| Area | What to check |
|---|---|
| Business discovery | Businesses in ChatGPT is a beta for claiming and managing a business profile. Claiming does not guarantee referral traffic. |
| Personal finance and apps | Paul reports ChatGPT finance access on more surfaces, a Notion editor in ChatGPT, multi-machine Dots and Claude Duet session control. Check your account's availability and read/write permissions before moving a sensitive workflow. |
| Smaller models | Meta's Muse Ember targets repeated, token-heavy agent tasks; its efficiency claims come from Meta's selected tests. |
| Other agent access | The episode mentions Muse outbound business calls and Grok Bot connections through Tailscale. Test geographic eligibility, identity, network permissions and approval steps before enabling them. |
| Google surfaces | Antigravity's knowledge-work layout, Gemini in CarPlay, memory summaries and Gemini glasses are discussed in the video. Treat the glasses' quoted price and availability as an episode snapshot; consult the current product page before buying. |
Video Chapters
| Time | Segment |
|---|---|
| 00:00 | Intro |
| 00:59 | NVIDIA CEO resigns |
| 02:46 | Fable and Carbon |
| 05:40 | Gemini 4 Hydrogen |
| 06:48 | Choosing your AI |
| 09:13 | Kortex MCP demo (sponsor) |
| 10:54 | Claude office suite |
| 16:23 | Grok Bot 1.0 |
| 18:00 | AI on your wrist |
| 21:21 | Rapid-fire updates |
| 26:24 | Final thoughts |
Sources and Caveats
Prepared from the episode and supplied transcript, and the linked primary NVIDIA, Anthropic, Google, xAI, OpenAI, Meta and Kortex pages. Checked 13 November 2026. The Kortex portion is sponsored in the original video; this article is not sponsored. Paul's model preferences and watch benefits are his experience. Provider benchmarks, prices, plan limits and rollout details can change. The NVIDIA investigation should not be interpreted as proof of a security breach.