Key Terminology

This glossary defines the vocabulary used throughout the book — both terms we coined and technical terms we borrowed from AI and software practice. If a term stopped you mid-sentence, flip here.

Terms marked with (Compound) are specific to the framework described in this book. Terms marked with (Technical) are industry-standard vocabulary explained for non-technical readers.

Compound Framework Terms

AI Readiness Scorecard (Compound): A five-dimension diagnostic tool — Constraint Clarity, Information Readiness, Workflow Visibility, Decision Rights, and Measurement Discipline — scored Red/Yellow/Green to show a leadership team exactly where their AI readiness gaps are before starting a Sprint. (First introduced: Chapter 1)
At-Risk Source (Compound): Institutional knowledge that lives in one person’s head and is about to leave the organization through departure, retirement, or role change. Named and prioritized during Source so it can be captured before it walks out the door. (First introduced: Chapter 6)
Build Spec (Compound): The eight-section developer-ready specification produced in Build, derived from the Design Brief. The first three sections (workflow summary, inputs, outputs) and the systems and integrations section are restatements of Design Brief content at developer-execution depth. Four sections add what Design doesn’t own: agent scope at the call level, human supervisor role, failure-mode behavior, and build path with implementation constraints. If the spec disagrees with the Design Brief, the Design Brief wins. Produced via the Build Spec Writer instrument. (First introduced: Chapter 8)
Center of Excellence (COE) Operating Model (Compound): The organizational design pattern that coordinates specialized functions across a large organization through defined governance, accountability, and recurring alignment sessions. Originates in Dave Ulrich’s three-pillar HR framework — HR Business Partners, Centers of Excellence, Shared Services — from Human Resource Champions (1997). The Compound quarterly operating session borrows the COE governance discipline (fixed agenda, committed outputs, recurring cadence) and adds the four-step structure: review the previous Sprint, update the Constraint Backlog, commit the next Sprint, update the Hybrid Accountability Chart. (First introduced: Chapter 11)
Clarity Call (Compound): A 30-minute, no-cost conversation with a Compound strategist designed to pressure-test a constraint and determine whether it is Sprint-ready. Not a sales pitch; the output is a named constraint you can act on. (First introduced: Chapter 12)
Co-Intelligence (Compound): Human intelligence and machine intelligence working the same problem. Neither one is enough alone; the combination is the part that compounds. The first term in the equation Co-Intelligence + Co-Operation + Rhythm = Compound. (First introduced: Chapter 3)
Co-Intelligent Co-Operation (Compound): The activity of humans and AI working the same problems together, inside the same loop, against the same outcome — as the company’s default way of working rather than a project or initiative. (First introduced: Chapter 1)
Co-Intelligent Company (Compound): What a company becomes when humans and AI work as one system, not two parallel tracks. A higher bar than “uses AI tools” or “has an AI strategy.” (First introduced: Chapter 1)
Co-Operating Model (Compound): The explicit design of who brings what intelligence to the work and who owns which operating responsibility — the structure that makes Co-Intelligent Co-Operation possible. The answer to the operating-model problem most companies have. (First introduced: Chapter 3)
Co-Operation (Compound): How the work is split between humans and agents: humans own outcomes, agents own tasks. The deliberate design of who does what, who reviews, and where work moves between them. The second term in the equation Co-Intelligence + Co-Operation + Rhythm = Compound. (First introduced: Chapter 3)
Compound (stage) (Compound): The sixth and final stage of the Sequence. After a Sprint is delivered, Compound runs the Sprint Retrospective, re-ranks the Constraint Backlog, installs one design change, updates the Hybrid Accountability Chart, and names the next constraint — turning a shipped project into the foundation of the next Sprint. (First introduced: Chapter 4)
Compound Bench (Compound): Six AI coaching agents — one per stage of the Sequence — that run the framework with you during a Sprint: interrogating constraints, surfacing knowledge gaps, running Work Deconstruction, specifying builds, structuring rollouts, and closing the retrospective. Available through Compound membership.
Compound Coach (Compound): A Compound strategist who works with members through weekly coaching calls — helping close gaps, resolve uncertainties, and guide Sprint execution. Remote, not on-site. The human support complement to the Compound Bench’s agent support.
Compound Sprint (Compound): One complete pass through the six-stage Sequence (Signal → Source → Design → Build → Deliver → Compound) on one validated constraint. The unit of work this book teaches. (First introduced: Chapter 4)
Compounding Scorecard (Compound): A running table that tracks the cost delta for each Sprint across quarters — constraint solved, cost before, cost after, delta, and cumulative impact — so leadership can see the compounding returns building over time. (First introduced: Chapter 11)
Constraint Backlog (Compound): The standing inventory of validated candidate constraints the organization is working through — born in Signal (Chapter 5), where the first constraint is named and quantified. Re-ranked during every Compound session as new information emerges; reviewed in Rhythm each quarter. The top entry becomes the input for the next Sprint’s Signal conversation. Selection is load-bearing-first: the governing constraint whose removal unblocks the others. Cost serves as the inventory view and breaks ties, not as the primary selection criterion. (First introduced: Chapter 5)
Constraint Statement (Compound): The one-page artifact that Signal produces: a single sentence naming the operational constraint, plus the location in the org, its duration, the quantified cost, and the validating evidence. The handoff from Signal to Source. (First introduced: Chapter 5)
Decision Rights (Compound): One of the five dimensions of the AI Readiness Scorecard. The degree to which a leadership team has explicitly assigned which tasks AI handles versus which stay with humans, and who has authority to change the boundary. (First introduced: Chapter 1)
Design Brief (Compound): The single artifact Design produces. It captures every decision Design has committed to before prototyping and Build: workflow summary, stakeholders, systems involved, data requirements, success criteria, constraints and guardrails, and V1 artifact description. It’s the sole upstream input to the Build Spec, and the source of truth on what the system should do and why. (First introduced: Chapter 7b)
Design Gate (Compound): The five-item checklist that must be fully checked before a Sprint moves from Design into Build: Work Deconstruction complete, all HAC fields filled with named supervisors, Human Orchestrator named, AI-Assisted vs. Automated decided with rationale, and guardrails defined. (First introduced: Chapter 7b)
Diagnose / Execute & Compound (Compound): The two halves of the Sequence. Diagnose = Signal + Source (the work most companies skip). Execute & Compound = Design + Build + Deliver + Compound (ships only when Diagnose is solid). (First introduced: Chapter 4)
Human Orchestrator (Compound): The operator who runs the redesigned workflow and supervises its agent team day-to-day once it ships. An existing employee whose role the Sprint upgrades, trained during the Sprint to direct the agents, review their output, and handle exceptions. Distinct from the leadership sponsor (accountable for the outcome) and the Sprint Lead (who runs the Sprint project). The detailed treatment of the role is in the Designing the System chapter. (First introduced: the Framework chapter)
Hybrid Accountability Chart (HAC) (Compound): An extension of a traditional accountability chart that includes agent teams alongside humans. Every row names a role or function, the agent team responsible, the human supervisor, and the AI-Assisted vs. Automated level. Every row has a named supervisor — no exceptions. A single Sprint can produce multiple HAC entries. (First introduced: Chapter 7)
Hybrid Org Today (Compound): A one-page living document that gives the current operational picture of the Co-Intelligent Company: active Sprint status, Constraint Backlog ranked by cost, Hybrid Accountability Chart entries, completed Sprint outcomes, and the next quarterly review date. Updated during each quarterly operating session. (First introduced: Chapter 11)
Information Flow Specification (Compound): A spec for how information moves through one accountability — naming inputs, outputs, transformations, decision points, and handoffs between humans and agents. Distinct from the Knowledge Map (which is the atlas of what’s known); the information flow spec is the route the work takes. (First introduced: Chapter 7)
Knowledge Map (Compound): The one-page Source deliverable that shows what the organization knows about the validated constraint — every relevant data source, its owner, its status (Clean/Needs Work/Missing), and its pipeline connectivity. Includes Digital sources, Organic sources (people), and Missing sources. (First introduced: Chapter 6)
Meridian Manufacturing (Compound): The fictional through-line case study used throughout this book — a $7.2M custom metal fabrication shop in Grand Rapids, Michigan, with 27 employees and a quoting bottleneck costing $558K/year. The company, its characters, and its Sprint are illustrative composites, not a real company. (First introduced: Case Studies)
Organizational Memory (Compound): The layer of connected knowledge — CRM records, SOPs, transcripts, client history — that makes an agent “know” a company rather than just the world. Built deliberately through Source and maintained Sprint by Sprint. Prevents the loss of institutional knowledge when people leave. (First introduced: Chapter 3)
Parallel Workstreams (Compound): Multiple agent-driven work tracks running simultaneously under one human’s direction. The design pattern that collapses calendar time and produces compounding output — contrasted with linear work, where tasks run sequentially because all workers are human. (First introduced: Chapter 3)
PIS Framework (Compound): The three-phase diagnostic structure — Problem, Identify, Solution — that prevents organizations from moving to solution before the problem is fully understood. The Compound Sprint embeds PIS into its order: Signal is the Problem phase (name and quantify the constraint), Source is the Identify phase (map what the organization knows about the constraint), and Design through Deliver is the Solution phase. (First introduced: Chapter 6)
Rhythm (Compound): The quarterly cadence of running Sprints. What turns a single Sprint into a compounding business outcome. Running the Sequence over and over — with a quarterly operating session to review the previous Sprint, update the Constraint Backlog, commit the next Sprint, and update the HAC. (First introduced: Chapter 4)
Right Seat Evaluation (Compound): The three-test discipline for assigning a human supervisor to every Hybrid Accountability Chart row: Sees It (does the supervisor understand the agent’s work well enough to evaluate output quality?), Wants It (are they genuinely accountable for the outcome, not just named on the chart?), and Suited for It (do they have the judgment, context, and authority to override the agent and make the calls the agent can’t?). A supervisor candidate who fails any test is not a supervisor — choose someone else or build the development plan that gets them to all three. (First introduced: Chapter 7)
Sequence (Compound): The ordered six stages inside the Framework: Signal → Source → Design → Build → Deliver → Compound. Always all six. Always in that order, because each stage produces something the next stage requires. (First introduced: Chapter 4)
Signal (stage) (Compound): Stage 1 of the Sequence. The discipline of identifying the one operational constraint worth solving, tracing symptoms to root cause, and quantifying the cost in dollars or hours before committing a Sprint to it. Produces a one-page Constraint Statement. (First introduced: Chapter 4)
Source (stage) (Compound): Stage 2 of the Sequence. Maps the information environment around the validated constraint — what knowledge exists, where it lives, who holds it, what’s missing, and what’s at risk of leaving. Produces the Knowledge Map. (First introduced: Chapter 4)
Sprint Lead (Compound): The person who runs the Sprint project day-to-day — cadence, handoffs, scope. Who fills it varies: often the leadership sponsor in a small company, a delegated lead or an outside coach in a larger one. Distinct from the Human Orchestrator, who operates the shipped workflow, and the leadership sponsor, who’s accountable for the outcome. (First introduced: the Framework chapter)
Sprint Planning Canvas (Compound): A one-page artifact completed before Signal begins: names the constraint, maps a first-pass answer to each stage, identifies the team, sets the Human Orchestrator, and books the quarterly review. Intentionally incomplete — Signal will refine it. (First introduced: Chapter 4)
Sprint Retrospective (Compound): The structured review that runs inside the Compound stage: what worked, what didn’t (with each “didn’t work” reframed as a design question), and what one design change would make the next Sprint better. Run with the Human Orchestrator and all Sprint accountability owners — the people who operated the shipped workflow. (First introduced: Chapter 10)
TML Framework (Compound): The three-layer classification system — Task, Management, Leadership — that separates organizational work and knowledge by the type of intelligence required. Task knowledge and work follow defined rules and can be fully systematized (Fully Automatable). Management requires judgment about process and is best handled with human-AI collaboration (AI-Assisted). Leadership requires judgment about people, strategy, and context and must remain with humans (Human Judgment Required). Source uses TML to categorize what the organization knows (Chapter 6); Design uses TML to categorize what the work requires (Chapter 7b). (First introduced: Chapter 6)
Work Deconstruction (Compound): The Design-stage instrument that classifies every task in a constraint workflow into the three categories of the TML framework — Task (Fully Automatable), Management (AI-Assisted), Leadership (Human Judgment Required) — to determine what the agent team handles and what stays with a human. (First introduced: Chapter 7b)

Technical Terms

API (Application Programming Interface) (Technical): How software systems talk to each other. Think of a waiter carrying orders between a kitchen and a table: the API carries requests from one system and brings structured answers back from another. When a system “has an API,” other software can query it and get data in return without anyone typing it manually.
Backlog (Technical): A prioritized list of work items to be completed. In software development this is a task list for a development team; in the Compound Sprint context, the Constraint Backlog is the prioritized list of operational constraints the company is working through.
Chunking (Technical): How documents are broken into smaller pieces for storage and retrieval in a RAG system. The size and structure of chunks affects how accurately the AI finds and uses information when answering questions against a knowledge base.
Connector (Technical): A pre-built integration between two specific software systems. Platforms like Zapier, Make, and n8n offer thousands of connectors that link one tool to another without writing custom code — for example, automatically creating a project record in a PM tool when a deal closes in a CRM.
Context Engineering (Technical): The practice of designing what information an AI model has access to and how it’s structured — context windows, system prompts, project instructions, knowledge bases, and RAG configurations. Distinct from prompt engineering, which focuses on the instruction itself. (First introduced: Appendix)
Context Window (Technical): The maximum amount of text an AI model can “see” at one time — the combined length of what you send it plus what it generates back. Longer context windows let the model work with larger documents or longer conversations before losing track of earlier content.
CRM (Customer Relationship Management) (Technical): Software that stores and manages a company’s relationships with customers and prospects — contact records, deal history, communication logs. Examples: HubSpot, Salesforce.
Data Pipeline (Technical): An automated flow that moves data from one system to another, often transforming or processing it along the way. A data pipeline might pull records from a CRM, run them through an AI model, and write the results to a spreadsheet — without anyone doing it manually.
Embeddings (Technical): A way of representing words, sentences, or documents as lists of numbers so a computer can measure how similar two pieces of text are. Used in vector databases and RAG systems to find the most relevant documents for a given question.
ERP (Enterprise Resource Planning) (Technical): Software that manages a company’s core business processes — inventory, finance, supply chain, manufacturing. Examples: SAP, NetSuite, QuickBooks Enterprise.
ETL (Extract, Transform, Load) (Technical): A data process that pulls data from one system (Extract), reshapes it into the right format (Transform), and stores it in another system (Load). Common in data engineering when connecting systems that store information differently.
Fine-tuning (Technical): Retraining an AI model on your company’s own data so it behaves differently by default — changing how the model “thinks.” Expensive, slow to set up, and hard to update. Appropriate only when you need fundamentally different default behavior across thousands of interactions. Contrast with RAG, which is faster and more flexible for most business use cases.
Hallucination (Technical): When an AI model produces a confident, plausible-sounding answer that is factually wrong. The model generates text based on patterns, not verified facts, so it can “make up” details — especially when asked about specific facts, names, or figures it doesn’t have reliable data on. A core reason human review of agent output matters.
Knowledge Base (Technical): A structured collection of documents, articles, SOPs, or records that an AI can search and reference when answering questions. The company’s institutional knowledge made accessible to agents through a retrieval system.
Large Language Model (LLM) (Technical): The AI technology underneath tools like Claude, ChatGPT, and Gemini. Trained on enormous amounts of text, an LLM learns to recognize and generate language at scale. The same underlying technology powers both simple chatbots and sophisticated agent workflows — what differs is the architecture built around it.
Low-code / No-code (Technical): Software development approaches that let non-developers build functional workflows and applications using visual interfaces, drag-and-drop tools, and pre-built components rather than writing code from scratch. Examples: Zapier, Make, Airtable, n8n, Power Automate.
MCP (Model Context Protocol) (Technical): How an AI model connects to your company’s tools and live data. An API lets systems talk to systems; an MCP lets an AI model talk to your systems — reaching into your CRM, reading your knowledge base, or checking a project management tool in real time during a workflow. Without MCPs, an AI model only knows what you paste into the prompt.
Metadata (Technical): Data about data. In a knowledge base or document library, metadata tags each document with information like date, author, topic, or status — so retrieval systems can find the right documents and know whether they’re current or archived.
Open Source (model) (Technical): AI models whose underlying code and weights are publicly available, allowing companies to run them on their own servers rather than sending data to a provider’s API. Examples: Llama (Meta), Mistral. The key advantage is data control; the trade-off is that someone must maintain the infrastructure.
Orchestration (agent) (Technical): Coordinating multiple AI agents so they work together on a task — passing outputs from one agent to another, managing sequencing, handling errors. The Human Orchestrator oversees orchestration within the Compound Sprint.
Prompt (Technical): The instruction you give an AI model — the text that tells it what to do, in what format, with what constraints. Prompt quality directly affects output quality. A well-constructed prompt includes the agent’s role, what it receives as input, what it should produce as output, and the constraints and edge cases it should handle.
Prompt Engineering (Technical): The skill of writing precise, testable instructions for an AI model so it reliably produces the output you need. Most useful once you know what the agent is accountable for — which is the output of Signal, Source, and Design.
Proprietary Model (Technical): AI models built and hosted by a company (Anthropic, OpenAI, Google) and accessed via their API. You send data to their servers; they process it and return results. Most capable models are proprietary. The trade-off versus open source is that data travels through the provider’s systems.
QA (Quality Assurance) (Technical): Testing a built workflow against real inputs to confirm it produces the expected outputs, handles edge cases correctly, and escalates appropriately — before declaring it done. In the Compound Sprint, this is the “Done test” in Build: five questions answered against last week’s actual data.
RAG (Retrieval-Augmented Generation) (Technical): Giving an AI model access to your company’s specific documents and data at the moment it needs them, without retraining the model itself. Think of handing the model a reference binder before each question. The model stays general-purpose; your knowledge provides the specifics. Faster and cheaper to set up than fine-tuning, and easy to update as your information changes.
Serverless / Edge Deployment (Technical): A way to run software (including AI workflows) on distributed infrastructure that scales automatically based on demand, without managing dedicated servers. Useful for high-volume, unpredictable workloads.
SOP (Standard Operating Procedure) (Technical): A documented, step-by-step description of how a process is performed. A well-written SOP is the foundation for an AI agent that executes that process — it is the knowledge the agent needs to do the work consistently.
Structured vs. Unstructured Data (Technical): Structured data lives in defined fields — database columns, spreadsheet rows, CRM records. AI can use it immediately. Unstructured data is free-form — emails, PDFs, meeting transcripts, Slack messages. AI needs processing and classification before it can reliably reason against it.
Swim Lane Diagram (Technical): A visual workflow map divided into horizontal or vertical lanes, one per actor (human or agent), showing which steps each party owns and where handoffs cross between them. Used in Design to make the workflow legible and spot fragile handoffs before they become problems in Build.
Token (Technical): The basic unit AI models use to process text — roughly three-quarters of a word. Every prompt you send and every response generated consumes tokens, and providers charge per token. A 1,000-word document is approximately 1,300 tokens. Understanding token volume matters for estimating the cost of production AI workflows.
Vector Database (Technical): A type of database that stores information as embeddings (numerical representations) rather than rows and columns, allowing fast similarity searches — “find me documents most related to this question.” Used in RAG systems so agents can quickly retrieve relevant knowledge from large document collections.
Webhook (Technical): A mechanism that lets one software system automatically notify another when something happens — for example, when a form is submitted, a payment is received, or a status changes. Webhooks are the “if this, then that” triggers that power many workflow automations without manual intervention.