Andrew Warner and Corey Ganim cover 15 AI releases and stories in one fast episode. The pattern is bigger than any one launch: personal agents are competing to become useful coworkers, while smaller products are packaging one narrow task well enough to earn a place in a real workflow. This guide puts the original sites, repositories and verified posts next to each item, with a practical check before you spend time or money.
Start with agents if you need delegated research or repeatable work; try creative tools if you need a concrete deliverable; check email and infrastructure if reliability is the bottleneck. Claude Duet is still a limited research preview. The PHP warning is urgent for operators: verify and patch the runtime, not just the CMS or app code. Product claims and changing revenue dashboards are not independent performance tests.
Watch the Episode
The video is presented by Zapier. Use the chapter table to jump straight to a launch; links and availability below were checked on 22 October 2026.
Agents That Can Keep Working
These five stories ask the same business question from different angles: can an agent hold context, work in the right environment, and produce a result a human can inspect? The answer depends on access, permissions and team structure, not just the model name.
1. Claude Duet
What the episode shows
A proactive personal agent that works in connected apps and brings back briefs. Andrew and Corey compare it with ChatGPT Dots and Grok Bot, then argue that one shared agent is less useful to a team than several role-specific workers.
Practical check
Anthropic calls this a US-only Pro/Max research preview on a waitlist; Team and Enterprise are not included yet. It shares Claude usage limits and is designed to ask before consequential actions. The hosts did not get to run a full team test.
2. Hyperagent for Agencies
What the episode shows
Andrew describes separate agents for sales, operations and channel growth, shared by teammates in Slack. The agency offering lets a service provider deploy and maintain agent teams for multiple clients.
Practical check
Test client isolation, approval roles, usage billing and what persists after an agency handoff. Discounts and performance anecdotes in a launch post are not guarantees for another client.
3. Ask an Agent to Find Savings
What the episode shows
The hosts suggest auditing subscriptions and bills, then asking where a business is spending effort without a return. Corey says the revenue-opportunity question is the more interesting test for an already lean business.
Practical check
Start with read-only statements and an explicit date range. Verify every proposed cancellation, fee change or revenue idea against the underlying records; do not let an agent terminate services or move money from a hunch.
4. PhoneUse
What the episode shows
PhoneUse offers agents a cloud phone they can tap and swipe, opening workflows that are easier in a mobile app than a web browser or API.
Practical check
Confirm device, account and data isolation before adding a personal login. An agent using a phone still has to follow the app's terms and ask before purchases, messages or sensitive account changes.
5. Grok Bot Pages
What the episode shows
The team highlights Grok Bot making reports, memos and dashboards as editable pages that can be revisited, not just one-off chat answers.
Practical check
Check the data source, refresh schedule and sharing permissions on any page. A visually polished dashboard is not evidence its numbers are current or its calculations are right.
Creative, Commerce and Design
These products put AI into a specific output: a gift, a visual interface, a video, a design canvas or research for a thumbnail. They are easier to judge when you compare the final artifact with your actual audience and rights requirements.
6. Pet Plus Us
What the episode shows
The offer combines a personalized portrait preview with paid digital or physical keepsakes. The business point is the finished gift, not a raw image-generation prompt.
Practical check
Compare the live price, likeness, shipping and refund terms before buying. TrustMRR is a changing revenue snapshot, not proof that a similar store will succeed.
7. OpenUI
What the episode shows
Thesys's open-source interface framework lets AI responses include interactive visual components, so an answer can become a chart, form or selectable view rather than a wall of text.
Practical check
Inspect the repo's supported framework and model integration before adopting it. Keep generated UI accessible, source-backed and explicit about any action it can take.
8. Pika in ChatGPT
What the episode shows
A Pika integration lets a user start a creative video workflow from ChatGPT. The hosts consider the appeal of moving from a product concept to a shareable clip in one conversation.
Practical check
Check current account access, credits, rights to uploaded images, render limits and how much editing remains possible before promising a production-ready ad.
9. MuseCourt
What the episode shows
A playful virtual courtroom where agents can argue a dispute. The hosts use it as a glimpse of agent-to-agent products, not a serious legal service.
Practical check
Treat the outcomes as simulation or entertainment. Do not put confidential disputes into a public service or treat an agent verdict as legal advice.
10. OpenDesign
What the episode shows
An open-source, collaborative design workspace for people and coding agents. The video positions it as a way to work on visual outputs together rather than passing static screenshots back and forth.
Practical check
Review local-versus-cloud storage, model keys, exports and team access before moving client designs into it. The repo and hosted app may have different cost and setup paths.
11. Thumbsy
What the episode shows
A searchable thumbnail research and comparison tool for YouTube creators. The practical use is finding patterns and previewing a design where viewers will actually see it.
Practical check
Use references to learn composition and contrast, not copy another creator's protected artwork. Test your own thumbnail against a representative audience and device size.
Email, Research and Infrastructure
The last group is less theatrical and more consequential. It includes two newsletter/email products, an Android TV utility and a PHP advisory. One of these could directly affect the security of a live site today.
12. MailCheap
What the episode shows
Marc Lou pitches a newsletter tool priced by emails sent rather than dormant subscribers; Andrew and Corey discuss the public argument with beehiiv. There is a hosted product and a separate self-hosted repository.
Practical check
Compare your own sending volume, deliverability, imports, automations and support. A low send price is not a total-cost comparison, and the open-source version still needs hosting, an email provider and maintenance.
13. Inbox Zero
What the episode shows
An open-source email assistant that can triage, label and automate inbox tasks. It is the operations counterpart to the week's flashier creative releases.
Practical check
Email access is highly sensitive. Review OAuth scopes, data retention, model providers and self-hosting requirements; start with labels or drafts before enabling send or delete actions.
14. CleanYourTV
What the episode shows
A developer uses Claude Code to make a utility for removing unwanted apps and changing settings on Android/Google TV devices.
Practical check
This is not a universal smart-TV tool. Check device compatibility, warranty and a restore path before disabling system packages; keep a record of every change.
15. PHP Security Advisory
What the episode shows
The final segment shifts from tools to two serious PHP remote-code-execution vulnerabilities. This is an operational update, not another product to try.
Practical check
PHP published fixed releases on 21 October: 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Check the actual web-serving runtime or vendor backport and update promptly. The existence of a flaw does not establish that every PHP site has been breached.
Original Posts and a Duet Hands-On
The X posts below are the full URLs we could verify from the truncated video-description links. They provide the original framing for Duet and the MailCheap/beehiiv discussion; they are not independent tests. Each has a direct link below its embed in case X cannot load in your browser.
For an actual Duet account walkthrough, Peter Yang shows what he could build and where the preview failed, including browser handoff. His experience is one creator's early-access test, not a universal product benchmark.
Three Small Tests Worth Running
- Run an agent value audit: list recurring software charges and one business metric; ask for source-linked opportunities, then personally verify one saving or offer change. Record the result against agent cost.
- Make one artifact: try one OpenUI answer, Pika clip, OpenDesign prototype or Thumbsy comparison using a real brief. Check export quality, accessibility, rights and whether the result saves work after revision.
- Protect the operational layer: inspect Inbox Zero email permissions and, if you operate a PHP site, check the deployed runtime against the fixed release list. Security review is not optional just because a tool is open source.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | 15 AI releases and tools to know |
| 00:25 | Claude Duet: Anthropic's agent |
| 01:02 | Duet versus ChatGPT and Grok Bot |
| 02:12 | Why businesses want specialized agents |
| 02:46 | Hyperagent for agencies |
| 03:40 | Agents finding savings and revenue |
| 05:04 | Pet Plus Us portraits |
| 07:31 | OpenUI visual responses |
| 08:53 | Pika inside ChatGPT |
| 10:07 | PhoneUse cloud phones |
| 12:46 | MuseCourt agent courtroom |
| 13:39 | OpenDesign collaboration |
| 14:46 | MailCheap versus beehiiv |
| 18:37 | Grok Bot reports and dashboards |
| 20:53 | Thumbsy thumbnail research |
| 21:35 | Inbox Zero email automation |
| 23:08 | CleanYourTV with Claude Code |
| 24:50 | Critical PHP vulnerabilities |
| 26:02 | Consequences for developers and services |
| 27:00 | Final thoughts |
Sources and Link Notes
Episode: Andrew Warner and Corey Ganim's video and the reader-supplied transcript. Primary records: linked product sites and repositories in each entry, Anthropic's Duet access guide, and PHP's 21 October security notice. The X post addresses supplied with the video were mostly shortened with ellipses. The three embedded status URLs and Ryan Doyle's linked TV demo were independently resolved; for other posts, this article links the relevant product, repository or public X account instead of guessing a status ID. Availability, plans, revenue snapshots and prices can change. Checked 22 October 2026.