proletariat

The Orchestration Playbook.

Mental models for putting AI agents to work.

Anthropic • Claude Code Community Meetup • Denver
Chris McDermut • February 2026

The Autonomy Ladder

Level 0 No AI "I don't use AI tools"

Level 1 Reference "I ask AI questions like a search engine"

Level 2 Copy-Paste "I chat with AI and paste code into my editor"

Level 3 Agent "AI edits my files directly"

        Level 4
        Parallelization
        "Multiple AI agents on different tasks"
      

        Level 5
        Directed Autonomy
        "I direct and review. Agents execute."
      

        Level 6
        Proactive Autonomy
        "Agents generate, review, and ship their own work"
      

Most teams are at Level 1–3. Orchestration enables 4–6.

Every System
Needs a Loop.

No loop = task / Loop = flow

A task list — pick up, do, deliver, repeat
A project board — tickets flow through statuses
A while loop — while(true) { work() }

The Work Matrix

Traditional — you write, team reviews
AI-assisted review — you write, agent validates
Supervised agents — agent writes, you review
Full autonomy — well-scoped work + automated review
No planning, no review — slop cannon

The Orchestration
Trilemma

Velocity + Delegation = speed

Velocity + Planning = control

Planning + Delegation = throughput

The fastest path is the one
you don't walk twice.

Orchestration
Primitives

The building blocks agents operate on.

Unit of work — ticket with spec, acceptance criteria, scope
Workspace — isolated git worktree per agent
Environment — container or VM the agent runs in
Permissions — what the agent can and can't do
Executors — orchestrator + agents that do the work
Gates — tickets in, PRs out, review at the boundary

The Orchestrator

PMO feeds Claude.
Claude delegates to the runtime.

Claude Code — orchestrator + agent runtime
Workspace — worktrees, containers, isolation, tools
PMO — tickets, specs, board state, status

PMO + Runtime + Claude = orchestrator

The Setup

Your terminal is the cockpit.

iTerm2 + tmux — split panes per agent, persistent sessions
Oh My Zsh + Powerlevel10k — branch, path, context at a glance
CLAUDE.md — project context every agent inherits
MCP servers — give Claude CLI tools it can use
Hooks, aliases + hotkeys — automation, shortcuts, fast navigation

Proletariat

npm install -g @proletariat/cli

Open-source orchestration that encodes these trade-offs.

The Loop — agents pick up tickets, plan, execute, deliver, repeat

The Matrix — configurable review gates per task type

Planning-first — agents plan before they code

Isolation — each agent gets its own branch + container

Requirements — specs, acceptance criteria, subtasks

Demo

Let's Connect

Star on GitHub

Download the CLI now

npm i -g @proletariat/cli

Chris McDermut • linkedin.com/in/christopher-mcdermut • github.com/chrismcdermut

The Orchestration Playbook.

Mental models for putting AI agents to work.

The Autonomy Ladder

Every SystemNeeds a Loop.

The Work Matrix

The OrchestrationTrilemma

OrchestrationPrimitives

The Orchestrator

The Setup

Proletariat

Open-source orchestration that encodes these trade-offs.

Demo

Let's Connect

Every System
Needs a Loop.

The Orchestration
Trilemma

Orchestration
Primitives