Miadi Review

[WILSON.ep120] Relational Multi-Agent Orchestration and Responsible Closure in Miadi Factory

Oct 13, 2026 · model: manual-import · channel: JGWill

Watch source

In the chronicle

[WILSON.ep120] Relational Multi-Agent Orchestration and Responsible Closure in Miadi Factory

Source: screenwalk-261013-research-is-ceremony-relational-ai-firekeeping-responsible-closure — JGWill
https://youtu.be/YOwqNj3jFmw
type: screenwalk
issues: sound cuts off at 70% , missing parts to reshoot or inject aomehow from jgi: https://chatgpt.com/s/t_6ace8700ca6c819192d202c43cab4f58
status: partial research material

Summary

Bridging Shawn Wilson's Indigenous research methodology with multi-agent systems, the speaker explores transforming agent pipelines from rigid, process-centric task execution into relational stewardship. Walking outdoors in the rain, he uses a mobile multi-modal voice loop to reflect on Sundance firekeeping teachings, agent session lifecycles, and why completing an orchestration cycle differs from merely optimizing efficiency.

Intended use:

redaction and help with ceremonial technology development

Academic Fields

Indigenous Research Methodology (Research Is Ceremony, Wilson 2008)

Wilson, Opaskwayak Cree, presents research as ceremony kept through relationality and relational accountability: the researcher answers to every relationship the work touches. The video carries this to agent teams and session closure. The 2008 book predates agentic software. That agents hold relations is the speaker's extension, not Wilson's claim.

Multi-Agent Systems (Computer Science)

This field studies autonomous agents that coordinate, delegate and share goals. Levesque, Cohen and Nunes's joint-intention theory holds that a team must track the status of the shared effort, which fits the video's handoffs and closure. These are formal models of agents with specified beliefs and commitments. A language-model session guarantees neither.

Human Oversight in Software Process (Software Engineering)

The firekeeper is a human steward over automated work. Bainbridge's "ironies of automation" set the limit: the more a system runs itself, the less practice the operator gets, yet the operator keeps the tasks nobody automated and must monitor for rare failures. Naming an accountable person does not make oversight effective.

Human–AI Interaction (Human–Computer Interaction)

This field studies people working with AI systems. Amershi and colleagues give 18 guidelines, grouped by when they apply, including when the system is wrong. The video's voice loop is one such system. They are heuristics, not proof a product works. The header's note that sound cuts off at 70% records a capture failure.

Provenance and Research Data Management (Information Science)

This field records data origin and movement. W3C PROV models entities, activities and agents; the video's handoffs, open dependencies and dispatch to collaborators are provenance questions. FAIR makes data reusable by machines but does not decide whether sharing is right. CARE adds collective benefit, authority to control, responsibility and ethics for Indigenous data.

Embodied Cognition and Walking (Cognitive Psychology)

Experiments test how bodily activity shapes thought. Oppezzo and Schwartz found walking raised idea generation over sitting, most outdoors, but not single-answer problem solving, in mostly student samples. Walking suits gathering ideas, not settling them, which fits the speaker's choice not to turn reflections into tasks early.

Aspects

Timecodes, the tools named (ElevenLabs, Scrivener) and the 100% figure below are as summarized from the video, not checked. YouTube lists the video at 45:13, so every timecode falls inside it. The header notes that sound cuts off at 70%. Against 45:13 that is about 31:39, so the third result (42:20-45:12) lies in the part without sound.

Research as Relational Emergence (00:18-01:34)

Rather than treating inquiry as a linear pipeline from fixed hypothesis to extracted answer, research emerges through sustained relationships. Listening and contextual resonance govern what information matters before formal questions are even framed [00:18-01:34].

The Firekeeper in Digital Systems (06:08-07:40)

Drawing on Sundance teachings, the speaker examines the firekeeper's role in digital ceremonies. Automation can handle scheduling and dependencies, but human stewardship remains accountable for maintaining respect, orientation, and relational integrity across the architecture [06:08-07:40, 23:20-24:40].

Relational Multi-Agent Architecture (18:24-20:30)

Moving beyond isolated agent sessions executing tasks independently, the Miadi Factory structures agents into teams defined by coordinated responsibilities. Inter-session relationships, handoffs, and shared context replace narrow execution benchmarks as the primary design principle [08:08-09:00, 18:24-20:30].

Embodied Mobile Inquiry (25:00-26:25)

Walking in the rain grounds thinking in physical environment, movement, and voice. Spontaneous audio recordings capture nascent observations without forcing premature execution or converting exploratory reflections straight into rigid tasks [11:25-11:53, 25:00-26:25].

End-results and Core Aspect

Core Aspect — Responsible Closure Over Task Completion (22:45-23:10)

A session lifecycle must distinguish technical artifact success from responsible closure. Like musical cadences or the medicine wheel, ending an agent cycle does not require 100% resolution; it explicitly tracks unfinished dependencies and acknowledged handoffs across cycles [10:15-10:45, 22:45-23:10].

Results

  • Relational Lifecycle Evaluation via Decision Engine (09:57-10:25): Jev is asked whether the criteria for responsible closure are satisfied across session webs [09:57-10:25, 22:53-23:10]. Jev returns a typed value with a probability, not a decision. Only a threshold applied in code, or a person's judgment, turns it into a closure decision. The summary does not show that this prevents uncoordinated termination.
  • Voice-to-Text Pipeline with Speech Synthesis (12:35-15:20): Field dictations are transcribed, synthesized into conversational audio personas via ElevenLabs, and verified on mobile devices while walking [12:35-15:20, 26:00-26:25].
  • Archival Integration and Collaborative Dispatch (42:20-45:12): Synthesized insights are formatted into Scrivener drafts, synced across platforms, and dispatched directly as research inquiries to academic collaborators [42:20-42:40, 44:45-45:12]. This section lies after the point where the header says sound cuts off, so the summary cannot rest on speech here. Not checked.

Sources