---
type: "raw_signal"
id: "the-thin-orchestrator-how-we-made-our-ai-0f3aa2"
slug: "the-thin-orchestrator-how-we-made-our-ai-0f3aa2"
title: "The Thin Orchestrator: How We Made Our AI Agents More Reliable by Giving the Router Less to Do"
description: "Meta published a production account of a “thin orchestrator” architecture for data-pipeline agents."
retrieval_nugget: "Meta published a production account of a “thin orchestrator” architecture for data-pipeline agents."
observed_at: "2026-10-20"
record_date: "2026-10-20"
date_kind: "observed_at"
why_it_matters: "Source: https://engineering.fb.com/2026/10/19/ml-applications/thin-orchestrator-ai-agent-architecture"
novelty: "notable"
verification_level: "source-linked"
signal_type: "research"
source_platform: "engineering.fb.com"
topics: []
entities: ["Meta"]
related_patterns: ["harness-architecture-outlives-model-choice"]
source_url: "https://engineering.fb.com/2026/10/19/ml-applications/thin-orchestrator-ai-agent-architecture"
source_urls: ["https://engineering.fb.com/2026/10/19/ml-applications/thin-orchestrator-ai-agent-architecture"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "signal:the-thin-orchestrator-how-we-made-our-ai-0f3aa2"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/signals/the-thin-orchestrator-how-we-made-our-ai-0f3aa2/","markdown":"https://newruntime.com/signals/the-thin-orchestrator-how-we-made-our-ai-0f3aa2.md","json":"https://newruntime.com/signals/the-thin-orchestrator-how-we-made-our-ai-0f3aa2.json"}
---

# The Thin Orchestrator: How We Made Our AI Agents More Reliable by Giving the Router Less to Do

## Retrieval answer

Meta published a production account of a “thin orchestrator” architecture for data-pipeline agents.

Meta published a production account of a “thin orchestrator” architecture for data-pipeline agents. The system uses a 1,300-line orchestrator to route more than 20 workflows to 14 specialized agents; the orchestrator performs no domain work. It handles intent classification, required-field gathering, reliability checks and delegation, while specialists hold isolated domain knowledge and perform the substantive reasoning.

Meta says configuration preparation for a new data column fell from three to four days by hand to under an hour with the system. The company also reports that about two-thirds of the system’s changes land without significant human edits. Workflow templates sequence the same pool of specialists for different processes, allowing a new pipeline-migration use case without new agent code.

The result is a first-party company case, not an independently evaluated benchmark. Meta’s evidence supports the architectural description and reported internal outcomes; it does not establish that the same gains will generalize to other organizations or workloads.

Source: https://engineering.fb.com/2026/10/19/ml-applications/thin-orchestrator-ai-agent-architecture

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- collection: [Raw Signals](https://newruntime.com/signals.json): Open every source-linked raw signal.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)