{
  "title": "Articles/every-app-becomes-an-agent",
  "caption": "Every App Becomes an Agent",
  "slug": "every-app-becomes-an-agent",
  "tags": [
    "agents",
    "article",
    "automatic-computer",
    "hermes-published",
    "pack-7",
    "published"
  ],
  "canonical_url": "https://mosiah.org/articles/every-app-becomes-an-agent/",
  "interactive_url": "https://mosiah.org/#Articles%2Fevery-app-becomes-an-agent",
  "markdown_url": "https://mosiah.org/articles/every-app-becomes-an-agent.md",
  "json_url": "https://mosiah.org/json/every-app-becomes-an-agent.json",
  "fields": {
    "sort-date": "2026-05-12T12:25:00Z",
    "caption": "Every App Becomes an Agent",
    "created": "20260512115443017",
    "modified": "20260512115443017",
    "tags": "article hermes-published published automatic-computer agents pack-7",
    "title": "Articles/every-app-becomes-an-agent",
    "type": "text/vnd.tiddlywiki"
  },
  "text": "//Related:// [[sources|Article Sources/every-app-becomes-an-agent]] · [[notes|Article Notes/every-app-becomes-an-agent]] · [[metadata|Article Metadata/every-app-becomes-an-agent]] · [[Published Pieces]]\n\n! Every App Becomes an Agent\n\n//The implementation layer becomes agentic. The user-facing ontology should not.//\n\nEvery serious app is becoming an agent.\n\nThis does not mean every app becomes a chatbot. It means every app gains the ability to observe state, interpret intent, make changes, run background work, call tools, verify results, and surface meaningful updates. The app stops being a passive interface over a database. It becomes an active process over a living object.\n\nA writing app becomes an editor, researcher, critic, citation assistant, publisher, and memory system. A video app becomes a producer, clip finder, captioner, timeline assistant, source manager, renderer, and version controller. A coding app becomes a refactorer, tester, reviewer, deployer, bug hunter, documentation writer, and architecture guard. A finance app becomes an analyst, reconciler, alert system, forecaster, and audit trail. A learning app becomes a tutor, curriculum mapper, question generator, misconception detector, and spaced-repetition engine.\n\nThe implementation layer becomes agentic.\n\nBut the user-facing ontology should not become agentic.\n\nUsers do not want to wake up to a dashboard of synthetic coworkers. They do not want to negotiate with a sales agent, writing agent, calendar agent, code agent, and research agent. They do not want personality management as a substitute for app management. They do not want the bureaucracy of the computer exposed as conversation.\n\nThe user still wants to work with objects: a draft, timeline, spreadsheet, codebase, room, board, source bundle, claim graph, project, playlist, dossier, page, video, or radio stream.\n\nAgents should be internal organs. They should not all introduce themselves.\n\nThis is the difference between an agentic app and an agent persona. An agent persona says “I am here to help you,” centers itself as a conversational partner, narrates, and simulates service. An agentic app simply changes state. It knows what object it owns, what invariants it must preserve, what tools it may call, what events it must record, and when the user needs interruption. It does not need to be loved. It needs to work.\n\nThe app becomes more like a living machine than a talking employee.\n\nPersonifying every software function creates a false social layer. Social interaction is expensive. It consumes attention. It invites performance. It makes users polite, annoyed, dependent, or impatient. It turns work into management.\n\nA good interface hides unnecessary agency. Nobody wants the spellchecker to have a personality. Nobody wants the garbage collector to introduce itself. Nobody wants the video renderer to express gratitude. We may want agents to explain themselves when something important happens, but the default should be quiet competence.\n\nThe agentic future should feel less like managing people and more like working with better instruments.\n\nThis is why durable artifacts matter. If the artifact is canonical, the agent can disappear into the work. A document has revisions. A codebase has diffs. A timeline has cuts. A claim graph has citations. A task has status. A workflow has events. The user sees the changed object and the evidence of change.\n\nIf the agent is canonical, everything becomes discourse. The agent says what it did, what it thinks, what it plans, what it recommends. The user must infer the state from the agent’s narration. That is fragile. Narration is not verification.\n\nEvery app becomes an agent, but the better app will not brag about it.\n\nIt will expose state. It will preserve history. It will let the user ask questions. It will let agents work in the background. It will make consequential changes inspectable. It will know when to interrupt. It will keep the user focused on the artifact.\n\nAgents behind the glass. Artifacts in front.\n"
}