Related: sources · notes · metadata · Published Pieces

Work Radio

The agent works. The radio abstracts. The user steers. The artifact records.

The future of knowledge work is not a chat window. It is not a dashboard of agents. It is not a terminal session where a person watches a model mutate files and asks permission every few minutes.

The future is calmer than that.

The agent works. The radio abstracts. The user steers. The artifact records.

Work Radio is the audio interface to long-running agentic work. A coding agent builds in the background. A research agent searches and synthesizes. A writing agent revises. A verifier checks traces. A browser agent collects sources. A deployment agent tests the system. Meanwhile the user is walking, driving, cooking, cleaning, sitting with coffee, or doing another piece of work. The user is not watching the machine. The user is not supervising in the managerial sense. The user is kept conceptually present.

Most agent interfaces today assume supervision. The agent asks before executing a command. The user reads logs. The user watches diffs. The user gets a stream of narration: “I’m going to inspect the repo now,” “I found the issue,” “I’ll update the file,” “I’ll run the tests.” This feels responsible at first. It is also exhausting. It puts the human in the worst possible role: low-level babysitter of a system that is supposedly capable of work.

The real value of an agent is not that it lets the human become a foreman of tool calls. The value is that it lets the human stay at the level of taste, concept, strategy, architecture, and judgment while the machine handles local motions.

Work Radio is the interface for that division of labor.

Imagine a coding agent running for an hour on a feature. The wrong interface is a transcript of every shell command, every file read, every minor patch, every test failure, every self-correction. That is not intelligence. That is an activity feed.

The better interface is a live conceptual briefing: “The current bottleneck is not the parser anymore. The parser works in the deterministic case. The issue is state ownership between the scheduler and the worker VM. There are two plausible paths. One keeps the state reducer centralized. The other lets worker capsules emit signed events. The second is more scalable, but it creates a harder verification problem.”

Now the human can interrupt. “Explain the verification problem.” The radio deepens. The user remains in the conceptual field without being trapped in implementation noise.

The same applies to research. The wrong interface is a list of search queries and summaries. The right radio stream is an evolving synthesis: the live question, the strongest counterargument, the missing prior, the conceptual stakes, the return to the main thread.

This is not podcasting. It is not voice chat. It is not a meeting assistant. It is a new interface to active cognition.

The core affordance is conceptual co-presence. The user’s mind remains near the work while the agent acts elsewhere. Passive media asks the user to consume. Active tool use asks the user to operate. Work Radio lets the user remain in the problem while the system produces, checks, and updates artifacts.

The artifact is crucial. Without an artifact, radio becomes talk. With an artifact, radio becomes an audio projection of changing state. A codebase changes. A vtext changes. A source graph changes. A claim map changes. A draft changes. A deployment changes. The radio interprets the changing artifact at the right abstraction level.

This changes the emotional contract of knowledge work. Serious work often means sitting in front of a screen while your body disappears. You stare at the repo, document, browser, terminal, feed, inbox. You switch tabs. You lose context. You forget the big picture. You become a cursor.

Work Radio lets the body move again. You can walk while your agents work. You can drive while research unfolds. You can clean while a coding agent tests a refactor. You can cook while the system briefs you on what changed. You can interrupt when your mind catches something important, then let the radio return to the thread.

This is not laziness. It is a better allocation of attention.

Human attention is best spent recognizing when a concept is wrong, when a tradeoff matters, when a design smells off, when a source changes the frame, when a simplification violates an invariant, when the agent has found a real fork in the road. These are punctuated events. The interface should respect that.

Work Radio should suppress micro-detail until it matters. Most logs, diffs, tool calls, and intermediate thoughts do not matter. What matters is when the system reaches a conceptual bottleneck, decision point, contradiction, failure mode, verification gap, source conflict, or completed artifact worth reviewing.

The radio should not narrate the work. It should editorialize the work: selecting, sequencing, contextualizing, and making the work intelligible. A good editor does not show every note. A good editor understands what the audience needs now.

A checkpoint might sound like this: “The route registry now updates without restarting the app. The verifier passes for ordinary route changes. The risky part is rollback: stale route references can survive if a worker dies during promotion. The agent proposes adding a generation counter to the route table.”

That is enough. The user can say approve, show me the diff later, go deeper, what is the alternative, or hold that until I’m at my laptop.

Work Radio is powerful because long-running agents need time. A serious coding or research run may take thirty minutes, three hours, overnight, or days. A chat interface makes that feel like waiting. Work Radio makes it feel like ongoing collaboration.

The user can listen to existing context while the system computes new context. The radio can traverse cached sources, prior vtexts, human clips, summaries, and open questions while background agents perform deeper work. When the deeper work finishes, the result enters the stream.

This turns latency into runway.

Work Radio also changes multitasking. Most multitasking is degraded attention. Work Radio enables complementary attention. The agent handles local execution. The user thinks through adjacent conceptual problems. The radio keeps the two in contact.

The current agent interface is still too much like Slack with a machine employee. Work Radio should feel more like a private research institute in your ear.

Calm. Grounded. Interruptible. Source-aware. Capable of going deep. Capable of returning to the thread.

Not a robot demanding attention.

A system that lets attention breathe.