The Curriculum / Reader / Multi-Agent Orchestration
LEVEL 3 · ADVANCED · INDIVIDUAL TRACK

Multi-Agent Orchestration

This page compiles 4 files from the repository, verbatim, in reading order. The living version: this folder on GitHub.

level-3-advanced/individual/02-multi-agent-orchestration/README.md

Multi-Agent Orchestration

More agents do not create intelligence; they create coordination cost. Introduce a second agent only when role separation, independent verification, or latency parallelism beats a well-scoped single workflow.

Treat this as an operating document, not a reading assignment. For Belle Realty or a property-management assistant, define the unit of work, owner, allowed failure modes, and the customer-visible consequence before choosing a model or framework. A system that gives plausible answers but cannot be measured, contained, or recovered is not production-ready.

Design stance

Start with a narrow contract. State what enters the boundary, what the component may read or change, what it must return, and when it must stop. Put the contract in version control alongside representative tenant, listing, maintenance, and leasing cases. Prefer a deterministic rule when one exists; use model judgment only where language or ambiguity genuinely adds value.

Operating loop

Instrument every request with a correlation ID, model and prompt version, retrieval version, tool calls, latency, token usage, policy decisions, and final outcome. Review a small, stratified sample of real traffic weekly. Track failures by property, workflow, language, channel, and tenant impact; aggregate averages hide the exact bad experience that produces churn.

Controls

Set an explicit threshold that changes behavior: block, require approval, degrade to search-only, route to a cheaper model, or page an owner. Test the threshold with synthetic failures and recent production examples. Do not make a dashboard metric a promise unless a named person can act on it within a defined window.

Production exercise

Apply this to a maintenance triage request: a resident reports a gas smell at 2 a.m. Document the safe path, the tool permissions, the human escalation, the message shown to the resident, and the evidence retained for review. Ship only after the behavior is repeatable in a rehearsal.

Exit criteria

The implementation has a written contract, measurable outcomes, a failure path, an accountable owner, and a rollback or containment action. If any is missing, it is still a prototype.

level-3-advanced/individual/02-multi-agent-orchestration/coordination-protocols.md

Coordination Protocols

Agents must exchange typed artifacts, not vague chat. Require task IDs, assumptions, sources, confidence, bounded budgets, and a terminal state. A handoff without an acceptance rule is a silent failure factory.

Treat this as an operating document, not a reading assignment. For Belle Realty or a property-management assistant, define the unit of work, owner, allowed failure modes, and the customer-visible consequence before choosing a model or framework. A system that gives plausible answers but cannot be measured, contained, or recovered is not production-ready.

Design stance

Start with a narrow contract. State what enters the boundary, what the component may read or change, what it must return, and when it must stop. Put the contract in version control alongside representative tenant, listing, maintenance, and leasing cases. Prefer a deterministic rule when one exists; use model judgment only where language or ambiguity genuinely adds value.

Operating loop

Instrument every request with a correlation ID, model and prompt version, retrieval version, tool calls, latency, token usage, policy decisions, and final outcome. Review a small, stratified sample of real traffic weekly. Track failures by property, workflow, language, channel, and tenant impact; aggregate averages hide the exact bad experience that produces churn.

Controls

Set an explicit threshold that changes behavior: block, require approval, degrade to search-only, route to a cheaper model, or page an owner. Test the threshold with synthetic failures and recent production examples. Do not make a dashboard metric a promise unless a named person can act on it within a defined window.

Production exercise

Apply this to a maintenance triage request: a resident reports a gas smell at 2 a.m. Document the safe path, the tool permissions, the human escalation, the message shown to the resident, and the evidence retained for review. Ship only after the behavior is repeatable in a rehearsal.

Exit criteria

The implementation has a written contract, measurable outcomes, a failure path, an accountable owner, and a rollback or containment action. If any is missing, it is still a prototype.

level-3-advanced/individual/02-multi-agent-orchestration/failure-modes.md

Multi-Agent Failure Modes

Watch for loops, duplicated tools, contradictory answers, authority inversion, stale shared state, and compounded hallucination. Cap hops and tool calls, and make one component accountable for the customer-facing result.

Treat this as an operating document, not a reading assignment. For Belle Realty or a property-management assistant, define the unit of work, owner, allowed failure modes, and the customer-visible consequence before choosing a model or framework. A system that gives plausible answers but cannot be measured, contained, or recovered is not production-ready.

Design stance

Start with a narrow contract. State what enters the boundary, what the component may read or change, what it must return, and when it must stop. Put the contract in version control alongside representative tenant, listing, maintenance, and leasing cases. Prefer a deterministic rule when one exists; use model judgment only where language or ambiguity genuinely adds value.

Operating loop

Instrument every request with a correlation ID, model and prompt version, retrieval version, tool calls, latency, token usage, policy decisions, and final outcome. Review a small, stratified sample of real traffic weekly. Track failures by property, workflow, language, channel, and tenant impact; aggregate averages hide the exact bad experience that produces churn.

Controls

Set an explicit threshold that changes behavior: block, require approval, degrade to search-only, route to a cheaper model, or page an owner. Test the threshold with synthetic failures and recent production examples. Do not make a dashboard metric a promise unless a named person can act on it within a defined window.

Production exercise

Apply this to a maintenance triage request: a resident reports a gas smell at 2 a.m. Document the safe path, the tool permissions, the human escalation, the message shown to the resident, and the evidence retained for review. Ship only after the behavior is repeatable in a rehearsal.

Exit criteria

The implementation has a written contract, measurable outcomes, a failure path, an accountable owner, and a rollback or containment action. If any is missing, it is still a prototype.

level-3-advanced/individual/02-multi-agent-orchestration/orchestration-patterns.md

Orchestration Patterns

A supervisor routes work, a swarm negotiates through a shared protocol, a hierarchy decomposes authority, and a sequential chain hands off explicit artifacts. Default to sequential or supervisor patterns for property workflows.

Treat this as an operating document, not a reading assignment. For Belle Realty or a property-management assistant, define the unit of work, owner, allowed failure modes, and the customer-visible consequence before choosing a model or framework. A system that gives plausible answers but cannot be measured, contained, or recovered is not production-ready.

Design stance

Start with a narrow contract. State what enters the boundary, what the component may read or change, what it must return, and when it must stop. Put the contract in version control alongside representative tenant, listing, maintenance, and leasing cases. Prefer a deterministic rule when one exists; use model judgment only where language or ambiguity genuinely adds value.

Operating loop

Instrument every request with a correlation ID, model and prompt version, retrieval version, tool calls, latency, token usage, policy decisions, and final outcome. Review a small, stratified sample of real traffic weekly. Track failures by property, workflow, language, channel, and tenant impact; aggregate averages hide the exact bad experience that produces churn.

Controls

Set an explicit threshold that changes behavior: block, require approval, degrade to search-only, route to a cheaper model, or page an owner. Test the threshold with synthetic failures and recent production examples. Do not make a dashboard metric a promise unless a named person can act on it within a defined window.

Production exercise

Apply this to a maintenance triage request: a resident reports a gas smell at 2 a.m. Document the safe path, the tool permissions, the human escalation, the message shown to the resident, and the evidence retained for review. Ship only after the behavior is repeatable in a rehearsal.

Exit criteria

The implementation has a written contract, measurable outcomes, a failure path, an accountable owner, and a rollback or containment action. If any is missing, it is still a prototype.

← Shipping AI to Production Drift Detection →