Paul J Lipsky's weekly AI news video has a useful split: new creative tools are getting easier to try, while always-on agents are getting harder to run without watching their allowance. The most immediately testable updates are ChatGPT Images 2.6 and Gemini Omni 1.2 Flash. Paul's best operational lesson is simpler: inspect what your agents repeatedly do before paying for a bigger plan.
Try one image edit or short video first, then check the output yourself. For agents, open the account's usage screen and review routine frequency before upgrading. Grok Bot's documented weekly pool is not the same as a published cost for each call or click. Meta's Muse Voice rollout starts in the US, and Claude Personal Finance is a US beta. Do not read a compelling demo as global access.
Watch Paul's Weekly Roundup
Creator: Paul J Lipsky. His demonstrations and reactions come from the video and supplied transcript. We checked product pages on 6 November 2026 for documented features and access; we did not reproduce his tests. The Bitwarden segment at 05:03 is sponsored.
What Changed This Week
| Update | What the source confirms | Reader check |
|---|---|---|
| ChatGPT Images 2.6 | Faster generation, improved edit consistency, GIFs and reusable reference images. The model is available across ChatGPT tiers; the two new features are rolling out. | Test one repeat edit. Check text, identity and details before sharing. |
| Gemini Omni 1.2 Flash | Up to 15-second video, conversational edits and API/business availability. Consumer app access varies by product and plan. | Google Vids has a limited free allowance; Gemini app access is rolling out to Ultra. |
| Grok Bot usage | A weekly usage pool and optional extra usage, depending on plan. Paul reports tighter limits in his account. | Read your own meter and top-up setting; do not infer a universal per-action price. |
| Muse Voice | Real-time voice and expressive avatars are rolling out in the Muse app, starting in the US. | Voice on glasses and Muse Charm is a future plan, not today's app feature. |
| Atmee V2 | The company offers a real-time avatar model, with V2 gated by a qualifying plan or invite code. | The striking video is a company demo, not an independent quality or cost test. |
Images and Video: Test the Edit, Not Only the First Render
At 00:31, Paul makes and edits images, tries GIFs and reuses a saved reference in ChatGPT. One caption in his demo does not land correctly, a good reminder to inspect the final file. OpenAI's launch claims roughly 50% faster image generation and distinguishes the API's Flare and Sunburst variants. That is a vendor speed claim, not a timed test by this site. OpenAI's release notes say GIF creation and saved reference images are rolling out, so their absence in one account is not necessarily an error.
Paul watches outside examples of Omni 1.2 at 02:38; he says he did not yet have access himself. Google's release says developers can use the API, while Google Vids offers up to three 15-second generations a month at no cost for Google account holders. That is an easier first test than buying access on the strength of a showcase clip. Review faces, logos, text, audio and usage rights before publishing generated media.
Voice Agents: Better Conversation, More Need for a Usage Budget
Paul's Muse Voice demo at 11:54 shows quick, interruptible conversation and selectable voices. Meta confirms a US-first rollout on mobile, web and Mac; its glasses and Charm plans are not current access paths. Paul likes the responsiveness, which is a personal assessment, not a head-to-head benchmark against ChatGPT or Grok Bot.
At 06:00, he describes Grok Bot limits feeling much tighter. The official FAQ describes a weekly pool shared with other eligible products under some plans and explains how extra usage is billed or capped. It does not provide a universal token bill for a browser action, phone call or routine. Paul's response at 07:11 is to review schedules, stop redundant checks, and use direct integrations where possible. The useful question is whether an unattended run changed a decision, not how many routines you can keep alive.
Paul also tests a Grok Bot outbound call by calling his own business. The Bot identifies itself as a virtual assistant, asks him for an answer it does not know, and shows a live transcript. That is a meaningful demo, but it is one call to a controlled recipient. Set an approved purpose and recipient, check the account's call controls, and have a human handle uncertain answers before using any calling workflow with customers.
The Other Updates: Access Matters More Than the Headline
Plans and work limits. Paul discusses Claude's Max/Ultra naming and ChatGPT/Codex allowances at 14:46 and 15:53. OpenAI's release notes confirm higher Work and Codex allowances for specified Pro tiers, not every subscriber. Check your own usage screen before assuming a limit change or a complete merger of Chat and Work pools.
Atmee and Suncatcher. The Atmee V2 clip is a vendor avatar demo; its model documentation limits V2 access by plan or invite. Google's Project Suncatcher update confirms a prompt was run on a prototype satellite and a response returned. It is an orbital hardware experiment, not a production space data center; Google itself identifies cooling as a major unsolved engineering challenge.
Personal finance and more agents. Paul mentions voice in ChatGPT Finances, new Dot models and Perplexity Teammates in his closing minutes. OpenAI documents Finances as an account-connected experience that cannot move money. Anthropic confirms Claude Personal Finance is read-only and rolling out to US Pro, Max and Ultra accounts; its numbers can be wrong, so verify important figures against the bank. We found no comparably clear primary product page for all the Teammates and Dots rollout details in this video, so treat those as Paul's report and check the current product screen before planning around them.
What Should You Try First?
| Your goal | First test | Success check |
|---|---|---|
| Create a reusable visual | Edit one ChatGPT image twice with the same saved reference. | Identity, text and layout survive both edits. |
| Make a short clip | Try one 15-second scene in Google Vids or the API you already use. | Useful final clip after review, within your actual allowance. |
| Keep an agent running | Audit one week's routines and usage before upgrading. | Same useful outputs with fewer unattended runs. |
| Try voice or finance | Check regional eligibility first; use a low-stakes question. | Correct answer, understandable permissions, and no assumed account access. |
Copyable Agent Usage Audit
This is an original JQ AI SYSTEMS prompt inspired by Paul's routine review, not a prompt from the video. It asks for a recommendation before changing an agent or payment setting.
Find the work worth keeping
Review routines before buying more capacity.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | This week in AI |
| 00:31 | ChatGPT Images 2.6 |
| 02:38 | Gemini Omni 1.2 Flash |
| 05:03 | Bitwarden sponsored segment |
| 06:00 | Grok Bot usage limits |
| 07:11 | How Paul saves Grok Bot usage |
| 10:21 | Meta Muse updates |
| 11:54 | Muse Voice hands-on demo |
| 14:46 | Claude Max and Ultra |
| 15:53 | ChatGPT and Codex limits |
| 18:23 | Grok Bot phone calls |
| 21:12 | Atmee V2 avatars |
| 23:41 | Project Suncatcher |
| 24:44 | ChatGPT Finances voice |
| 25:10 | Claude Personal Finance |
| 25:50 | ChatGPT Dots models |
| 26:04 | Perplexity Teammates |
| 27:28 | Paul's biggest takeaways |
Source Note and Product Links
Checked 6 November 2026. Paul's video and supplied transcript identify his demonstrations, opinions and chapter timings. Official product pages linked beside claims establish vendor-stated features and access. We did not independently reproduce his calls, image edits or voice comparisons, and availability, pricing and limits can change.
Create: OpenAI Images 2.6, Google Omni 1.2, Google Vids access. Agents: Grok Bot FAQ, Meta Muse Voice, Atmee models. Other: Claude Personal Finance, Google Suncatcher, ChatGPT release notes.