Listen to this podcast · 9:34
When Zuckerberg spoke with his Muse, Agrippa, at Connect, the scene recalled Tony Stark discussing a design with Jarvis. He asked for finishes for a small device and a revised arrangement of its components. Later, Andrew Bosworth gave his own agent a chain of design, scheduling and monitoring tasks. Their delivery presented conversation as an interface to work already moving underneath it. [1]stockanalysis.comMeta Connect 2026: Opening KeynoteThe speakers demonstrate Agrippa and Cooper, discuss the handheld project, and describe a future transaction-fee business model. Demonstrations and company claims remain attributed.Open source ↗
That resemblance matters. Fiction taught people to imagine a capable assistant that understood the project, kept track of unfinished work and acted while its owner continued thinking. The promise is that an idea can become something real while its author keeps thinking.
I read Meta's strategy through that promise. Glasses bring an agent into the immediate setting. A phone provides familiar access. A computer supplies an environment for detailed work. If Muse can maintain a useful relationship across those settings, Meta can become increasingly important even while someone continues buying another company's hardware.
The conversation must survive the next step
Meta describes Muse working in its own cloud virtual machine, using connected services and continuing after the user closes the application. It also describes remembering personal context and suggesting useful actions. Sensitive actions such as purchases require approval. [2]about.fb.comIntroducing Muse: The World’s First Personal AI Agent Built for EveryoneMeta describes Muse background tasks, availability and plans; advertising and confidential-VM statements retain their stated scope.Open source ↗
The important advance would lie in continuity. Consider a repair-shop owner who notices a recurring fault while examining a bicycle. In a plausible future workflow, the owner asks the agent to find the correct replacement, check availability and prepare an estimate. After permission, the agent orders the part, updates the customer and records the job. The owner keeps working.
A successful answer at the first step proves little about the rest. The agent must identify the correct item, preserve the budget, distinguish an estimate from an order and notice when delivery changes. Useful autonomy means carrying those obligations through the whole task.
That standard comes directly from my experience working with agents. A fluent explanation creates no value if I must reconstruct the assignment, repair the tool call and supervise every transition. I want an assistant that can recognize an ordinary obstacle, choose an authorized remedy and continue. I also want it to return when the decision genuinely belongs to me.
The voice matters because conversation lets people develop an idea before they know every step. An agent that works during that exchange could shorten the journey from an uncertain beginning to a finished object, a prepared meeting or a resolved problem.
Compass Predictive Analytics
September 27 evidence assessment: these instruments measure the qualified factual source corpus, not probabilities for the editorial scenarios.
Proactivity could return the time between tasks
The best opportunity may hide in work people never quite begin. An unanswered message, a renewal to compare, a promised introduction: each occupies a little attention while waiting for a convenient moment.
Imagine an independent filmmaker preparing a small production. During an evening conversation, the filmmaker discusses a location, a possible prop and a concern about the schedule. By morning, an agent has prepared availability questions, organized alternatives and identified the decision that prevents the rest from moving. The filmmaker begins with a clear next move instead of an expanding list of loose ends.
This is a scenario, not a claim about a complete shipping workflow. It illustrates why remembering context can matter more than producing another polished answer. Useful anticipation helps a person preserve momentum.
It also changes the meaning of convenience. Saving a minute on a search helps. Removing the need to remember that a search must happen can release a different kind of mental effort. That benefit becomes especially valuable to someone running a business, caring for relatives or balancing several responsibilities.
The risk follows the same mechanism. An agent that prepares the day's agenda helps determine what receives attention. Its suggestions must remain open to correction, including the correction that something does not deserve doing at all.
Compass Predictive Analytics
Three companies approach the same relationship
Apple describes photographers editing images in Pixelmator on MacBook Pro through Mac Virtual Display. [8]apple.comApple Vision Pro upgraded with the M5 chip and Dual Knit BandApple describes using Mac Virtual Display for professional work, including photo editing on a MacBook Pro.Open source ↗
That connection gives Apple an existing path between its spatial computer and familiar work. Tesla begins inside a vehicle people repeatedly use, with Optimus representing a prospective extension into physical work. Meta begins with social relationships and attention, then seeks a more direct role through Muse and wearable access. [3]tesla.comAI & RoboticsTesla describes fleet-scale AI and its Optimus development objective.Open source ↗
The Grok coffee demonstration I watched makes the connection tangible: a request becomes an order and a journey. The creator attributes the driving to FSD; the conversational agent and the driving system have different responsibilities. Tesla's documentation still identifies FSD as supervised. The strategic interest lies in coordinating capabilities through one understandable request. [4]youtube.comMy car drives itself and orders me coffee | Grok Bot in the TeslaCreator demonstration: order details, explicit authorization, collection reported, and FSD credited for driving. Auto-generated transcript inspected; not independent transaction verification.Open source ↗[5]tesla.comFull Self-Driving (Supervised)Tesla states that FSD requires active driver supervision and does not make the vehicle autonomous.Open source ↗
Muse could pursue the same advantage through a different setting: a conversation, something visible through glasses, or work waiting on a computer. Meta's existing relationships could help it find useful entry points. Apple and Tesla have their own recurring settings in which an agent can earn trust.
My prediction is that device loyalty and agent loyalty will begin to diverge. A person could carry an iPhone, use Grok in the car and rely on Muse for a continuing project. The company that carries the project well could become the first place that person turns, even when it does not own the screen.
That is why Apple must answer the agent relationship as well as the glasses. Better hardware alone cannot guarantee that people will choose its intelligence to organize their lives.
The small companion is an experiment in access
Muse Charm tests another route into that relationship. Zuckerberg's demonstration presented a small companion device that could travel on a keychain. I understand the appeal of an agent with a recognizable personality and an object of its own. I remain unconvinced that another thing to carry will become indispensable. [1]stockanalysis.comMeta Connect 2026: Opening KeynoteThe speakers demonstrate Agrippa and Cooper, discuss the handheld project, and describe a future transaction-fee business model. Demonstrations and company claims remain attributed.Open source ↗
The Tamagotchi comparison captures the emotional attraction while missing an important reversal. Here, the companion is supposed to reduce the owner's obligations. If charging, carrying and retrieving it add more friction than direct access removes, the form factor weakens its own purpose.
Glasses offer a stronger possibility when they leave the hands available and connect assistance to what the wearer is doing. A small dedicated device may still suit particular people or settings. Meta benefits from learning which form lets the relationship become useful with the least effort.
The durable product may therefore be the continuing agent, with several successful ways to reach it. No single accessory must become universal for that strategy to matter.
Compass Predictive Analytics
Capability can travel faster through experience
The Manus episode belongs in this analysis. Manus announced that it was joining Meta in December 2025 and later announced a return to independent operations. I infer that contact with an experienced agent team could have accelerated practical learning about tools, persistent work and recovery. The public record does not establish a particular transfer of code or architecture. [6]manus.imManus Joins Meta for Next Era of InnovationManus announced that it was joining Meta in December 2025.Open source ↗[7]manus.imManus Resumes Independent OperationsManus announced independent operations on September 1, 2026; the notice does not establish which capabilities may previously have transferred.Open source ↗
That distinction leaves the strategic hypothesis intact. Acquisitions, recruitment and internal research can compress the distance between a laboratory capability and a usable service. Meta's test is whether the resulting products keep improving once customers supply the complications a demonstration cannot anticipate.
The same test applies to every competitor. I would watch completed tasks, repeated use and the amount of supervision people still provide. An agent that asks intelligent questions and follows through can justify a place in daily life. A persuasive character that repeatedly loses the assignment cannot.
Compass Predictive Analytics
The promise is a more capable person
Picture a maker developing her first product after work. She brings the taste, experience and stubborn curiosity. The agent helps compare components, organize revisions and prepare the next experiment. A remote collaborator examines the design in a shared space. By the weekend, an idea has become something she can hold.
That is the Jarvis promise I find worth taking seriously. It extends the reach of a person's judgment. The work still needs an author, a purpose and someone who cares whether the result is good.
My expectation is that the first lasting gains will appear in these ordinary acts of creation and follow-through. They will accumulate through a better-prepared lesson, a repaired object, a small business that can serve another customer, or a project that finally escapes the notebook.
If Meta delivers that relationship at scale, the significance of Muse will exceed any one pair of glasses. More people could act on ideas they previously lacked the time, help or confidence to pursue. The invitation hidden inside the demonstration is a powerful one: bring the intention, and discover how much further you can take it.
About the evidence assessment
September 27, 2026. 3 qualified sources from 3 independently owned publishers. These instruments assess source-grounded factual assertions; editorial judgments and future scenarios carry no assigned probability. Two Tesla pages could not enter this metric corpus because retrieval was blocked.
