Layer L6

Orchestration

Workflow, routing, coordination. How skills compose into outcomes.

Why it matters

A single ring is useful. A curated jewelry collection with tasting, fitting, and custom design is an experience. Orchestration composes skills into workflows.

The Jewelry Store & Workshop

A single ring is useful. A curated collection with fitting and custom design is an experience. In AI: orchestration composes individual skills into multi-step workflows with human override and runtime assurance. One skill → one task. Orchestration → entire workflows.

The 5 sublayers

L6a

Agent Loops

Single-agent plan-act-observe cycles

L6b

Human-in-the-Loop

Escalation patterns, approval workflows, human override design

L6c

Role Routing & Task Decomposition

Breaking complex work into subtasks and assigning to the right agent

L6d

Context & State Management

Maintaining working memory, session state, context windows across steps

L6e

Runtime Assurance & Learning Loops

Post-deployment monitoring, evals, feedback pipelines, drift detection

, Layer diagnostic card · SCOI v1

Is a company really at L6?

Workflow coordination, how multiple skills, tools, and humans compose into a multi-step outcome.

Inclusion tests · include if ALL

Coordinates multi-step plans with branching, retries, human-in-the-loop escalation.
Manages cross-step context/state, not a single prompt-response.
Owns the runtime that decides which agent/tool/human handles which step.

Exclusion tests · exclude if ANY

Single prompt with a long context window, that's L2/L5, not L6.
A pipeline DSL with no runtime governance, closer to L4.
DAG editor with no L5 skills underneath, UX over emptiness.

The L6 removal test

Remove L6 and either the human stitches the steps together, or it collapses into a single L5 call. Either way, the 'orchestration' was overhead, not value.

Economic work this layer does

Converts a set of capable-but-isolated skills into a reliable end-to-end workflow with audit and human override.

Canonical examples

Glean
L1 corpus + L6 retrieval+routing + L8 enterprise memory. L6 is load-bearing.
Notion AI
L6 + L8 inside a distribution surface the user already opens daily.
LangChain (the framework)
Reference L6 primitives, but as a product, prone to becoming a feature.

Anti-examples · look-alikes that fail

Pure 'agent framework' startups
L6 with no L5/L1/L8 underneath. By Law I, absorbed by L2 platforms.
Zapier-style automators (AI-painted)
L4+L6 with no L8 memory, wins on inertia, vulnerable to embedded copilots.
Most "multi-agent" demos
Orchestration theater. No durable buyer outcome.

Disagree with a classification?Open the classification table →

Who's playing here

LangChainCrewAIZapier (at risk)Make (at risk)

Verdict: Contested. Becoming a feature, not a product.

Case studies touching L6

From Dashboard to Skill Hire: The Death of Per-Seat Software

Software has moved through five distinct eras of human–machine division of labor. We are mid-transition between Era 3 (The Dialogue, human directs, AI builds) and Era 4 (The Workspace, AI orchestrates, human supervises). Era 5 (The Skill Hire, the agent IS the worker) arrives by 2028. Per-seat pricing is structurally dead in Eras 4–5 because the seat itself goes away. Every product roadmap needs to be re-priced and re-architected along both the customer axis and the depth axis.

McKinsey + OpenAI (Lilli): When the Consulting Firm Owns the Memory, Not the Model

McKinsey didn't build a model. It built Lilli, an internal assistant trained on 100,000+ McKinsey documents, 70 years of proprietary studies, and the firm's named expert network. OpenAI provides L2. McKinsey owns L1 (the IP) and L8 (the firm's institutional memory). The consultant doesn't get disrupted by the model, the consultant rents the model and keeps the moat.

Cursor at $9B: The IDE That Quietly Became the Most Important L4 in AI

Cursor isn't a model. It isn't an agent. It's the editor, the place developers spend 8 hours a day. By owning L4a (IDE distribution) and layering L6c (agent orchestration) and L8d (per-codebase memory) on top, Cursor has become structurally more defensible than the agents that run inside it.

Glean at $7.2B: The Enterprise Memory Layer Microsoft Was Supposed to Own

Glean indexes every document, message, ticket, and meeting inside a company, then makes it queryable by AI. That index is L1c behavioral data (proprietary to each customer), the orchestration is L6d context management across apps, and the cross-app memory is L8d institutional knowledge. Microsoft 'should' own this with Copilot. They don't, and Glean's $7.2B valuation says the market noticed.

L5 Execution L7 Surface