Skip to main content

Case Study

Directors Chair: Mission Control for an Autonomous Agent Fleet

A real-time mission control dashboard monitoring 12 integrated agents across job search, project maintenance, and infrastructure.

PublishedMar 2026 - PresentArchitect & Operator
12 Agents IntegratedLive DashboardMother Hen Framework2-Machine Collaboration

What It Is

Directors Chair is a real-time mission control dashboard monitoring an autonomous agent fleet across job search, project maintenance, and infrastructure. Built on top of my Mother Hen framework, 12 integrated agents handle the operational work of finding, evaluating, preparing for, and tracking opportunities, while I stay in the loop on every decision that matters. The meta-angle is intentional: the system I built to find work is itself the strongest evidence of the work I do. The product is the portfolio.

The Lineage

Directors Chair did not appear from nothing. It sits at the end of an iterative engineering arc, where each project extracted and generalized patterns from the one before it.

1

MCP-WP

A LangGraph architectural prototype for WordPress analysis with 27 agent personas built on 7 agent types. The first system where multi-agent coordination patterns were developed and tested.

2

Ora

An orchestration framework extracted from MCP-WP. Generalized the coordination patterns into a reusable layer for managing multiple AI tools in parallel.

3

Mother Hen

A mission control framework built on Next.js, Supabase, Qdrant, and Discord. Provides the infrastructure for running specialized agent fleets against structured pipelines. Currently in private beta.

4

Directors Chair

The current instance of Mother Hen, configured for job-search operations. The live deployment running the pipeline described on this page.

Architecture

The system is layered. Directors Chair is the application instance: it defines the pipeline stages, agent roles, and domain-specific logic for job-search operations. Underneath, Mother Hen provides the framework: agent lifecycle management, pipeline orchestration, persistent state, and communication channels.

┌─────────────────────────────────────────────┐
│         Directors Chair (instance)           │
│   Pipeline: Discover → Score → Prep →       │
│             Apply → Follow-up               │
├─────────────────────────────────────────────┤
│           Mother Hen (framework)             │
│   Agent lifecycle · Pipeline engine ·        │
│   State management · 2FA auth               │
├──────────┬──────────┬───────────────────────┤
│ Supabase │  Qdrant  │   Discord (comms)     │
│  (data)  │ (vector) │   (notifications)     │
├──────────┴──────────┴───────────────────────┤
│        Specialized Agent Fleet               │
│   Discovery · Scoring · Research ·           │
│   Prep · Tracking · Follow-up               │
├─────────────────────────────────────────────┤
│           External APIs                      │
│   Job boards · Company data · Comms          │
└─────────────────────────────────────────────┘

Core Stack

  • Next.js ... dashboard UI, server actions, real-time views
  • Supabase ... pipeline state, opportunity records, application tracking
  • Qdrant ... vector storage for semantic search across opportunities and context
  • Discord ... notification channel for agent activity and operator alerts
  • Claude API ... powers the specialized agents across the pipeline

The Fleet

Twelve agents, three lanes. Pipeline runs the discover → score → prep → apply → follow-up workflow; maintenance keeps the public site and shipped projects honest; infrastructure keeps the fleet itself running. Each node pulses on synthetic activity — the real dashboard streams the same shape from Supabase. Click a node to inspect its role.

The Pipeline

Every opportunity flows through five stages. Agents handle the operational work at each stage; the human operator reviews, approves, and makes the judgment calls.

1

Discover

Agents monitor job boards, company pages, and referral channels to surface roles that match the target profile. New leads flow into Supabase as structured records.

2

Score

Each opportunity runs through a relevance-scoring pass: role fit, company signals, location, and timeline. Low-confidence leads are deprioritized automatically.

3

Prep

Scored leads trigger research agents that pull company context, recent news, and team information. The output feeds tailored application materials.

4

Apply

Application materials are assembled and staged for human review. Nothing submits without operator approval: the system drafts, the human decides.

5

Follow-up

Active applications are tracked across stages. Agents flag when follow-ups are due, surface context for upcoming conversations, and log outcomes back into the pipeline.

5

Pipeline Stages

Live

Dashboard Status

2

Operator Machines

Supabase

Pipeline Backend

Live Dashboard

Directors Chair is not a concept or a coming-soon page. It is running now. The dashboard at directors-chair-jj.vercel.app shows the current state of the pipeline: active opportunities, agent activity, and pipeline progression. What you see there is the system I use daily to manage this search.

What Runs Where

The system operates across a two-account, two-machine collaboration pattern. My account handles the operator role: reviewing agent output, approving applications, making strategic decisions. A collaborator (Henry) runs parallel agent sessions on a second machine, handling research-intensive and bulk-processing tasks. Both accounts connect to the same Supabase backend, so pipeline state is always consistent regardless of which machine made the last update.

This is not distributed computing for scale. It is a practical pattern for splitting operator attention: one person drives the pipeline, the other runs deep-research agents, and the database keeps them synchronized. The pattern works because the framework was built for it: Mother Hen manages agent sessions per-operator, with shared state and isolated execution.

Honest Tradeoffs

Directors Chair is built on my own daily-driver infrastructure. It ships when I ship. The pipeline quality depends directly on how much time I invest in tuning agents, refining scoring criteria, and reviewing outputs. There is no team behind this, no QA department, no SRE on-call. When it breaks, I fix it. The meta-irony is acknowledged: I am using the system to find work, which means the system's effectiveness is tested in real time, with real stakes. That is either the strongest possible validation or the most stressful possible QA environment, depending on the day.

April 2026 . -JJ

Tech Stack

Frontend

Next.jsTypeScript

Backend

Supabase

AI

QdrantClaude API

Tooling

Discord

Infrastructure

Vercel

Want to discuss the architecture?

Get in Touch
Directors Chair: Mission Control for an Autonomous Agent Fleet — Jeff Michael Johnson