Agent Orchestration Map
A visual map of how AIOS work is routed, reviewed, gated, and improved using fallback-safe role labels. This is a documented role-and-boundary map, not runtime execution proof; availability is static unless a matching worker receipt exists.
Count summary
Human lead
Sararin
Gate ownerAI Workforce Source Status
Verified by workspace registry
No global count shownBig Crew
Bounded specialist delivery workspace
Started — explicit handoff and separate receiptsSupernova
Business intelligence workspace
Bounded canonical candidate; POC and R01 start unproven.Investment Team
parked / TBD / not countable
Future lane1. Who does what
Agent and role lanes show who owns routing, execution, evidence, and gates.
Sararin
Sets the release decision and human boundary.
Final approvalGPT
Classifies work, checks claims, and keeps scope bounded.
Context and reviewHermes
Packages handoffs and keeps the worker lane clear.
Stage managementCodex
Changes files, runs checks, and reports evidence.
Scoped executionResearcher
Finds sources, receipts, and limits for claims.
Evidence scanSimple Task Supporter
Handles simple support work only after boundary confirmation.
Contract confirmation pendingEngineering / Quality role group
Engineering review workspace with explicit handoff and separate receipts; it is not an autonomous team claim.
Started — bounded specialist deliveryBusiness / Opportunity Intelligence workspace
Frames opportunities, risks, options, and business evidence; candidate personas are not active operational roles.
Bounded canonical candidateInvestment Team
Acknowledged future lane with no countable registry yet.
TBD / not countable2. How work moves
Example task enters AIOS, gets classified, packaged, executed, reviewed, gated, and converted into a learning artifact.
Sararin / GPT
Example task enters AIOS with purpose, risk, and expected output.
IntentGPT
GPT identifies task type, risk level, and safest lane.
LaneHermes
Work is prepared as a bounded handoff.
PacketCodex / Big Crew / Supernova / Researcher / Simple Task Supporter
The selected worker lane creates the artifact or review input.
OutputExecutor / Researcher
Receipts, sources, validation notes, or limits are captured.
TraceGPT / Reviewer
Output is checked for DoD, claims, and boundary risk.
DoDSararin
Sararin approves, revises, rejects, or parks the work.
GateGPT / Codex
Work is committed, published, revised, stopped, or archived.
StateGPT / KB
Lesson becomes checklist, prompt, or role rule update.
RuleSupernova role registry and POC boundary
Supernova: first version complete; POC not yet validated.
Opportunity Framer
Decision frame
Market & Pain Reality Analyst
Pain evidence
Brand & Positioning Strategist
Claim boundaries
Business Model & ROI Strategist
Economics view
Budget & Resource Optimizer
Resource guardrail
Marketing & Sales Channel Strategist
Channel test
Finance / Tax / Legal Gatekeeper
Risk flags
KPI & Metrics Analyst
Metrics threshold
Strategy Options Advisor
Option tradeoffs
Risk & Execution PM
Execution plan
3. How control and learning work
Evidence, review, Sararin Gate, rework, enforcement rules, and improvement outputs keep the map bounded.
Evidence gate
Executor / Researcher
Receipts, artifacts, validation, and source notes are captured.
Review gate
GPT / Reviewer
Scope, claims, Definition of Done, and risk are checked.
Sararin Gate
Sararin
Sararin approves, revises, rejects, or parks the work.
Rework loop
GPT / Worker lane
Feedback is classified, revised, checked, and captured.
Improvement output
GPT / KB
The lesson becomes a checklist, prompt, role contract, or routing rule.
Rework is treated as an evidence signal. Repeated feedback should become a checklist, prompt update, role contract, routing rule, stop condition, downgrade rule, KB note, or follow-up task. A rule only counts as improved when it becomes visible in a future check, route matrix, monitor status, or validation gate. This does not mean agents learn automatically.
- AIOS does not publish a global agent count until each workspace is backed by a committed registry or contract.
- Big Crew: started bounded specialist delivery with explicit handoff and separate receipts.
- Supernova: Stakeholder Core is a bounded canonical candidate; POC and R01 start are unproven.
- Simple Task Supporter: low-risk support lane, contract confirmation pending
- Investment Team: parked / TBD / not countable
- UX/UI: engineering capability tag, not separate role
- Big Crew role names remain hidden until source-of-truth confirmation is complete.
- Simple Task Supporter contract confirmation is pending.
- Investment Team is parked / TBD / not countable.
- UX/UI is a capability tag, not a separate role.
Big Crew source boundary
Started bounded specialist delivery; autonomous execution is not claimed.
Big Crew is a bounded specialist delivery workspace with explicit handoff and separate receipts; this does not claim an autonomous team, active roster, or self-validation.
Source: team-of-team-orchestration/role-output-contracts.md
Supernova source boundary
Stakeholder Core is a bounded canonical candidate.
The candidate defines stakeholder decision-support personas, not active operational roles. POC validation and R01 start are not yet authorized or proven.
Source: supernova/stakeholder-core/README.md + registry.json
Global count boundary
No global AI workforce count is shown.
The page avoids a single global agent count because some lanes are active, some are parked, and some are capability tags rather than standalone roles.
Source: team-of-team-orchestration/specs/aios-agent-review-dashboard-data-contract-v0.1.md
- unsupported counts
- role registry conflicts
- wrong worker/lane assignment
- missing evidence
- wrong Definition of Done delivery
- overclaim risk
- repeated rework patterns
- live telemetry
- autonomous orchestration
- real-time telemetry
- provider comparison
- benchmark performance
- cost-saving evidence
- production-grade observability
- final role-registry truth