Preprint Proposes Media as a Trigger for AI Agents: Audio Cues Route Playback to Workflow Queues

eTradeWire News/10853645
BOULDER, Colo. - eTradeWire -- Independent researcher Milan Cheeks has published a preprint, "Harmony: Media-Carried Workflow Signals for Distributed AI Agent Execution," proposing that music, video, and ambient audio can serve as an addressing layer for AI agent systems.

AI agents today start through apps, APIs, or consoles. Harmony embeds a compact packet, HMY2, into a media file's audio using near-ultrasonic frequency-shift keying. The packet carries no fixed task. It carries a routing identity: an organization, a workflow, a channel, and an agent suite.

An opt-in listening node decodes the cue locally — raw audio never leaves the device — and presents the routing fields to a runtime, which assigns the next approved task from the queue. No agent runs without explicit human approval, and each run is recorded in a receipt. Because the cue is a route rather than a payload, media authored once can address work that changes over time without re-encoding the asset.

The paper describes a complete six-component prototype — encoder, inspector, verification tools, listening node, reliability bench, and runtime — with the full loop demonstrated internally in June 2026. It is equally explicit about open work: real-world channel reliability is unmeasured, authentication is proof-of-concept grade, and the executor is template-based. A five-part validation program with acceptance criteria is set out in the paper.

More on eTradeWire News
Unlike the covert ultrasonic beacons criticized for tracking, Harmony requires opt-in listening, local decoding, readable and publicly inspectable cues, and human approval before execution.

"Media already reaches billions of playback events a day," Cheeks said. "Harmony asks whether that surface can address work, with consent and inspection built in."

The system is patent pending (U.S. Provisional No. 63/587,817, converted; held personally by the inventor). The preprint, not yet peer reviewed, is open access at doi.org/10.5281/zenodo.24380531 and https://www.milancheeks.com/research.html.

About: Milan Cheeks is an independent researcher in the M.S. in Computer Science program at the University of Colorado Boulder (ORCID 0009-0008-7507-0050). This research was conducted independently.

Secondary contact email: milan@selflabs.xyz

Contact
Milan Cheeks
mich3592@colorado.edu


Source: Milan Cheeks

Show All News | Disclaimer | Report Violation

0 Comments

Latest on eTradeWire News