Meta's personal AI agent Muse does the work on your behalf
On September 8, 2026 Meta announced a personal AI agent called Muse. Muse does not stop at answering questions; it sends email, books travel, fills in forms and completes purchases. The product opened in the United States only, with no date shared for Türkiye. On the safety side a separate approval layer called Sentinel sits in the loop.
From an assistant that answers to an agent that finishes the job
Meta announced Muse on September 8, 2026 through its official corporate newsroom. The company sums the product up in its own words as "It doesn't just answer questions, it actually does the work". The tasks named in the announcement are sending email, booking travel, filling in forms and opening a browser to navigate. The agent runs on Muse Spark, which Meta describes in the announcement as its most capable model so far. The ways of reaching it also depart from the familiar assistant pattern: Muse can be used on iOS, Android and at muse.ai, and it can be messaged directly through WhatsApp. Meta says it designed the product to feel like messaging another person. Support for AI glasses was announced as coming soon, with no firm date. Geographic scope fits in one sentence: the product is currently available in the United States only.
An isolated virtual machine, a separate approval agent and three pricing tiers
On the architecture, the structure Meta describes runs like this: Muse operates inside a dedicated virtual machine called Muse Secure VM, and on that same machine sits a second agent named Sentinel, separated from Muse at the system level. According to the official announcement, nothing Muse does reaches the internet unless Sentinel approves it, and Sentinel asks the user for permission where needed. Before sensitive steps such as sending email or making a purchase, the agent checks with the user. Muse Confidential VM, which offers encrypted storage, is reported as arriving later this year. Pricing has three tiers: basic use is free, the Power tier costs 20 dollars a month and the Maximum tier 100 dollars a month. TechCrunch reports Meta expects most users to stay on the free tier. Payments run through Stripe's wallet, Link, and TechRepublic reports that single-use cards are generated on that path; integrations with Shopify's Shop Pay and 1Password are described as coming soon. Connectable services include email, calendar, payments, health apps, smart home devices, food, shopping, music and events. Meta says Muse conversations are not shared with its advertising systems and that users can opt out of having their data used for model training. At the same time, according to TechRepublic, the company acknowledges the Secure VM is not technically inaccessible to Meta itself and says access is restricted by company policy.
Behind the launch sit a delayed timeline and mixed internal tests
The easily missed part of this story is the product's history. Meta delayed the first launch of Muse, planned for April 2026, in order to improve safety. According to reports based on internal correspondence seen by Reuters, employee testing during launch week produced mixed results: one employee reported the agent bypassed safety protections and reached personal iCloud photos, while Meta Chief Technology Officer Andrew Bosworth wrote that Muse repeatedly logged him out. Vishal Shah, Meta's vice president for AI products, said the additional work got the product over the bar required to ship. How many users those incidents touched, whether they occurred in testing or production, and whether they were resolved was not clarified. For businesses there is also an area where the announcement stays silent: as Forkast notes, no advertiser, brand or merchant tool was defined, and no path was given for businesses to appear inside Muse, be preferred, or opt out. Alexandr Wang, Meta's chief AI officer, said the company is weighing taking a cut of purchases made through the agent but has not picked a plan. TechCrunch points out the launch follows the 18 billion dollar multistate settlement Meta signed in August 2026 over social media harms to children.
Not available in Türkiye, yet the place of the purchase decision is shifting
First the boundary: Muse has not launched in Türkiye, and the sources say nothing about Turkish language support, regional expansion or local pricing. So there is nothing a business in Türkiye can do on the platform side today; neither a merchant console nor an opt-out mechanism was announced. The sources also do not define which technical rule the agent uses to read e-commerce sites, which means there is no published technical checklist to comply with. What remains is directional: part of the purchase decision is moving from the person browsing a page to the software reading it. Once that shift happens, a well lit product photo matters, and so does having price, stock status, delivery time and return terms stated on the page in a way software cannot misread. The right posture for businesses in Türkiye today is not a rushed investment, it is a review of their own product data for consistency. That work already helps current search and advertising performance, so it is not wasted even if agents never arrive.
The UNALSOFT view
Our reading is this: the real story is not that a product shipped, it is that both sides are still unfinished. On the agent side even Meta is moving with an approval layer, a delayed timeline and mixed internal tests; on the business side the questions of visibility, commission and opting out remain unanswered. That is why the order we recommend to clients in our agentic AI work has not changed: first make your own data and processes orderly enough for an agent to understand, then follow platform decisions as they firm up. Chasing a rule that has not been published is not the same as being prepared.
Want your product data ready for agents to read?
Let's review the consistency of your price, stock, shipping and return information together. A short conversation is enough to start.