Television (Agent GUI)

Television (television.run) is a Mac/browser application pitched as "the missing GUI for personal agents": a shared visual workspace where an agent pins its output as persistent artifacts — cards, interactive views, or external web pages — organized into channels, instead of everything scrolling away in chat. Any agent that can run commands and write files can drive it, and it ships shareable skills for building interfaces.

The page is a landing page, not an essay, so what it offers is a positioning claim more than an argument. That claim is nonetheless crisp and worth taking seriously: the chat transcript is a bad durable surface for agent work, and the fix is a visual space the human curates over time. The agent doesn't need a custom protocol — the bar is deliberately low ("if your agent can run commands and write files, it can use Television"), which makes Television a harness-adjacent product rather than a new agent: it assumes you already have one.

Key Quotes

"The missing GUI for personal agents."

Bold framing: it concedes the entire terminal-first world of coding agents and declares the next layer. The audience is explicitly "personal agents" — assistants and life/work automation — not software factories.

"Artifacts let your agent pin important information in Television instead of losing it in chat."

This is the core UX thesis and it matches a real pain: agent output lives in an append-only scroll, and anything you want to keep has to be manually copied into files or notes. Making the agent write to a pinned, structured surface inverts the flow — the transcript becomes the log, the board becomes the product.

"Organize your artifacts into channels: persistent spaces where you and your agent can collect, arrange, and return to the things you're working on."

Channels are the memory architecture in disguise: a persistent, human-arranged organization of agent output is a shared external memory between person and agent — closer to a wiki than a chat.

"Supercharge your agent with skills: reusable instructions for creating beautiful interfaces that you can adapt and share."

The skills system is the same pattern the coding-agent ecosystem standardized — reusable instruction bundles — applied to interface-building, so the agent's presentation layer is itself a curated, shareable artifact.

Key Themes

#tool — Television itself: a GUI for existing agents, Mac app plus browser fallback. #pattern — artifacts-and-channels as the durable-surface pattern replacing chat scroll. #concept — the agent-agnostic integration model: any filesystem-and-commands agent qualifies. #concept — interface-building skills, sharing the skill format conventions of the broader agent ecosystem.

Critical Analysis

The strongest idea here is the two-surface model: chat for the conversation, channels for the durable result. That is genuinely different from most agent GUIs, which are timeline viewers — pi-gui resurfaces the session transcript as a structured review surface; Television resurfaces the outputs as artifacts. Both reject the raw scrollback as the product; they diverge on what deserves the surface.

The marketing does real work but also hides the interesting questions. How does the agent write an artifact — a file convention, a CLI, an MCP tool? Nothing is specified, and that protocol is the actual product boundary. What's the permission model when an agent can write into a space the human treats as their own curated workspace? And is "artifacts as external web pages" an escape hatch or a prompt-injection surface — a persistent page the agent controls is a channel of influence that lives longer than any single message.

There's also an unacknowledged tension with memory systems: channels-as-organized-artifacts overlap heavily with what agent memory architectures (Agent Memory) try to do inside the model. Television externalizes that to the human's arrangement instincts — bet on taste over retrieval.

Relations

This strengthens Personal Agents as a topic: it's a pure play on the human-facing layer of a one-person agent stack, and it names the missing piece ("GUI") that most personal-agent projects leave as a chat window. It nuances pi-gui — both are desktop GUIs for existing agents, but pi-gui surfaces the process (timeline, sessions) while Television surfaces the output (artifacts, channels); the comparison sharpens what "agent GUI" can mean. It connects to Agent Memory by offering a human-curated external memory architecture — organized artifacts instead of retrieval — a competing answer to the same "agent output should persist and be findable" problem. And it echoes Intent Is the Interface's concern with what the human actually looks at when working with an agent: Television bets that surface is a board, not a log.

Sources: raw/television-run, Television Last updated: 2026-10-05