An ordinary user opens Meta Muse on a phone or iPad and hands it to-dos, queries or errands; the personal AI agent performs actions and returns results. The public material does not say which materials it handles, how far it delivers, or which steps still need human confirmation. Judgement: its significance is not the feature list but the entry point — when a platform puts a personal agent in a device's default position, general errands become infrastructure rather than a product. What is missing is the actual range of actions it can perform and its permission boundary.
Mac users run AdsNotch while working and hand over their screen including the notch area as ad inventory, earning from impressions. How advertisers buy, how it is priced and settled, and what AI actually does in it are all unstated in the public material. Judgement: treating a personal device interface as sellable inventory is a coherent direction, but the risk sits on whether advertisers really want this inventory and whether platform rules permit it.
Creators doing talking-head or explainer content record audio and gather footage, then pick a motion template and style in adu-motion-video and let Codex or Claude Code cut the voice-over and footage into a finished, publishable video. Judgement: the entry point for editing is shifting from timeline software to "templates plus a coding assistant", templating motion work that used to need manual keyframes — a fine-grained job platforms will not do themselves.
When developers run Claude Code and Codex at the same time, AgentDock collects and answers the prompts each agent raises from one unified inbox, avoiding switching between terminal windows; the delivery is one completed agent confirmation or reply. Judgement: coding agents are moving from a single session to parallel multi-session runs, turning the human role into triage and confirmation; this spot will get crowded, but whoever builds the inbox first has a habit advantage.
When indie developers, SaaS or course teams need a product promo video, they hand a product name or URL to coding agents such as Cursor or Claude Code; the agent researches the product, drafts several style options to choose from, captures 4K real pages, and cuts to the beat of the music. Judgement: promo videos, once outsourced per video and priced by editing hours, are being split into research, capture, beat-cutting, voice-over and review steps that can be automated, and pricing will follow.
An AI glasses brand for everyday wear, letting users handle voice interaction, capture and information viewing through the glasses while commuting, in meetings or on the move. The candidate material only discloses funding and valuation; the hardware form, AI feature boundary and delivery method still need verification. Judgement: AI glasses are moving from concept to a consumer hardware category with capital behind it and competition for wearable entry points is heating up, but hardware delivery and after-sales are the hardest part for an independent team.
Based on the available material, it targets people using AI agents and summons monster characters while the agent runs; what input it takes, whether it produces images or interactive characters, and what it ultimately delivers are not detailed publicly. Judgement: agent products are extending toward entertainment and companionship, turning tool calls into interactions with characters and emotional feedback, but the form is too unclear to conclude.
Before writing a design or running a review, a system architect feeds requirement documents and system notes in; the model reads them and produces an editable architecture diagram on a local Excalidraw canvas, and the person still has to check and adjust the shapes. Judgement: turning documents into diagrams is shifting from manual dragging to a model producing a first draft, and the local workspace is a key selling point for teams unwilling to upload internal documents to external services.