Andrew Warner's Muse review asks the right question: how much of a real job can an agent finish before it sends the work back to you? He watches creator demos and compares them with his own experience of phone calls, shopping, files, and messages. The useful pattern is clear. Muse can search, organize, draft, and monitor in the background. The less glamorous part is also clear: identity checks, uncertain recommendations, and repeated approvals can keep a task from being fully delegated.
Start Muse with information work: compare prices, extract a schedule from a document, or prepare a briefing. Use it to investigate a bill before you call, but do not count a proposed discount as money saved until the account changes. Keep purchase, message, and account-permission approvals explicit. Judge each workflow by completed outcome and time saved, not by how impressive the demo looks.
Watch Andrew's Muse Review
Credit: Andrew Warner and The Next New Thing's original video, published 28 September 2026. The results below are attributed to the creators shown in that video or to Andrew's personal tests. They are not controlled product benchmarks.
Make Money or Cut Costs: Check the Final Step
Negotiate a bill by phone
Peter Yang shows Muse calling a provider, disclosing that it is an assistant and returning a call transcript with a possible lower rate. Andrew notices the catch: the provider still wants the account holder to call and complete verification. This is a useful research assistant, not a confirmed autonomous saving. The test is the next bill, not the quoted offer.
In Paul J Lipsky's phone test, Muse identifies itself as AI and asks a business for its hours. Andrew says phone calling was still a limited rollout in his experience, and businesses may decline to talk to an agent. Ask for a narrow, legitimate business inquiry; do not assume voice access or acceptance is universal.
Track prices, then approve the purchase
The Tech Girl's demo turns a wish list into a tracker and proactive price alerts. This is promising because the agent can watch a market without a daily manual check. For a meaningful discount, record the exact item, baseline price, shipping, taxes, and alert threshold; compare against the live seller page before ordering.
Meta says Muse can check out with Link by Stripe and a one-time card that hides the underlying card number, and that Muse asks before purchases. That is an important protection, but it does not validate the product, merchant, delivery terms, or return policy. Andrew sees the approval step as friction; for payments it is also a deliberate control. Keep a modest spending cap for a first trial.
Browse Facebook Marketplace
Peter demonstrates searches, listings, and seller messages. Andrew's own bed-frame search was less successful: Muse suggested poor options for his needs, and he disliked confirming each outbound message. The right first test is a bounded search with dimensions, location, condition, pickup constraints, and a budget. Review the listing and seller yourself before messaging or paying. Faster browsing is not the same as a completed purchase.
Run the Household Without Losing the Source
School PDFs. In The Tech Girl's example, Muse extracts dates and policies from a school document so a parent can ask follow-up questions later. Andrew's improvement is to keep the important facts in a document or database the family controls, rather than relying on one agent's memory. Check any holiday, absence policy, or contact against the original PDF before acting. A child's records are sensitive: share the minimum needed.
Goals. Alex Finn shows Muse Goals prompting for context and proposing steps toward a creator-business objective. Andrew likes the guided questions more than unsolicited suggestions. Keep the goal specific: one outcome, deadline, constraints, and a weekly review. A generated LinkedIn carousel is a possible output, not proof the business goal was met.
Local files. Paul demonstrates reading and editing a file on a Mac. This makes a read-only summary an excellent first task. For edits, duplicate the file, name the expected change, and inspect a diff or reopen the document before trusting the result. Check the Mac app's current access settings before granting control of files or other apps.
Family briefing. The Tech Girl asks for a printable morning brief combining calendars, email, activities, weather, and local news. Andrew values the scannable design. This is useful only if the source data is current and the brief does not expose private details to the wrong audience. A first version can use a single calendar and one-page output before connecting the entire inbox.
Research, Then Decide What to Send
Browser research. Edwin Chen's demo compares voice-AI suppliers in a virtual browser and stops before payment. Meta says Muse works in a dedicated cloud VM with a browser. Use a source table with current price, plan conditions, geography, and date checked; confirm figures on the supplier's checkout page. In the video, signup and a bot challenge still require a handoff.
Messaging integrations. Peter and Andrew use Muse to triage messages across Meta apps. Andrew notes an important boundary: chatting with Muse through WhatsApp is not the same as giving it full management of your WhatsApp conversations. His own test of inbox monitoring surfaced little beyond spam, and he found confirmation for every outbound message cumbersome. Read-only summaries and draft replies are a sensible first step; human review is appropriate before sending anything in your name.
Andrew also mentions Zapier MCP in a promotional segment. Zapier's documentation describes app- and action-level permissions and action history. It is a separate integration route, not proof that every Zapier action is available natively in Muse. Restrict any agent to the specific apps and actions its task requires.
What the Demos Do Not Prove
- Research is not completion. A negotiated rate, Marketplace lead, or supplier comparison may still need identity verification, checkout, or human judgment.
- Availability varies. Meta initially rolled Muse out in the US on mobile and web. Andrew reports phone-call access was limited when he recorded; check your own account rather than assuming every demo feature is present.
- Approvals have a cost and a purpose. They slow rapid back-and-forth, but Meta says sensitive sends and purchases require approval. Count the interruptions, then decide which actions you actually want to delegate.
- Free has a boundary. Meta says most everyday use is free and offers subscription plans for more. Check current limits before designing a daily operation around it.
- Connected data deserves scrutiny. Meta says users can choose permissions, disconnect apps, and opt out of model-training use. Review those controls, the audit trail, and retention expectations before sharing family, financial, or client material.
A First-Week Muse Test
- Pick one low-risk queue. Use a shopping watch list, one school document with unnecessary personal data removed, or a public supplier comparison.
- Define the deliverable. Ask for a dated source list, a price alert with a threshold, or a one-page brief. Tell Muse to stop before sending, buying, or changing account settings.
- Limit access. Connect only the app or document required; prefer read-only access when it exists. Keep payment and outbound messaging behind explicit approval.
- Run it three times. Record elapsed time, number of corrections and approval prompts, freshness of sources, and whether a person still had to finish the task.
- Keep or cut. Continue only if the result is reliably useful, cheaper in attention than doing it yourself, and easy to audit. Remove unused permissions afterward.
Watch the Eight Source Demos
These are the videos Andrew uses to illustrate Muse. The creator tests show individual experiences, not guarantees that a feature is available in every account. The Meta keynote provides launch context.
Meta Connect 2026: product direction
Peter Yang: bills, shopping, and Meta apps
Paul J Lipsky: calls and local files
The Tech Girl: trackers and family briefs
The Tech Girl: school PDFs and memory
Alex Finn: goals
How I AI: agent experience
Edwin Chen: browser research
Andrew's Video Chapters
| Time | Section |
|---|---|
| 00:00 | Meet Meta Muse |
| 00:29 | Negotiating bills |
| 03:13 | AI phone calls |
| 05:26 | Price tracking |
| 07:10 | Shopping and payments |
| 09:01 | Zapier MCP |
| 09:53 | Facebook Marketplace |
| 11:56 | School PDFs |
| 13:20 | Goals |
| 14:47 | Local files |
| 17:41 | Family morning brief |
| 19:59 | Browser research |
| 22:24 | Messaging integrations |
| 24:51 | Andrew's verdict |
Sources and Credits
- Andrew Warner: Make money, cut costs & run your life with Meta's Muse (primary commentary and transcript).
- Meta: introducing Muse (availability, cloud VM, access controls, payments, and free/subscription positioning).
- Meta: Connect 2026 updates (Mac computer use, voice, and connectors).
- Zapier MCP (separate app/action permission layer mentioned in the video).
- Eight supplemental videos are embedded above with their sources and titles. The creator experiences are individual demos, not universal outcomes.
Video published 28 September 2026; article checked 29 September 2026. Product access, pricing, and connected-app permissions can change.