Agentic AI Glossary: Planning Terms Used in Canvas

1. 1Canvas and context
2. 2Readiness and value
3. 3Outputs for review

Completeness, readiness and value answer different questions.

This agentic AI glossary defines the planning terms used across Canvas: Agentic Brain, operating model, evidence, readiness and ROI confidence, in plain language. It covers this product's own vocabulary, not every agentic AI term. Keeping the terms distinct prevents the interface, documentation, and roadmap from making conflicting promises.

Agentic Brain

The Agentic Brain is the product name for the connected understanding the app builds about an AI opportunity. It brings together confirmed context, open questions, the operating model, readiness, memory, and the artifacts derived from them.

The Brain is not claimed to be conscious or independently knowledgeable. It becomes useful through the evidence and decisions accumulated with the user.

Operating model

The operating model is the precise current description of the operation the proposed AI will touch: people, systems, information, relationships, constraints, outcomes, and readiness. It is the structured entity underneath the product language of the Brain.

Canvas

The Canvas is the primary builder and editor for the operating model. Its areas guide discovery from business problem through systems, business impact, and an implementable workflow. It is not separate from the Brain; it is one of the main ways the Brain is built and corrected. The current Canvas has 11 authored sections. AI Readiness is computed from their evidence rather than authored as a separate section.

Evidence

Evidence is information supported by a user answer, attached source, public research, or an explicit decision. The origin and strength of evidence matter. AI-generated wording can organize evidence, but fluent wording does not make an unsupported claim true.

Starting situation

The starting situation is where the user says they are starting from, asked once at intake and changeable afterwards: shaping a first idea, restarting after a stalled or failed attempt, assessing gaps across teams, setting AI guardrails and governance, or evaluating a proposal or vendor pitch. This context helps shape questions and suggested next steps without hiding Canvas sections, readiness dimensions or outputs. You can also leave the starting situation unspecified.

Prior attempt

A prior attempt is a previous run at AI that the organization already made — a proof of concept, a demo, a vendor pilot, an internal build, or a project — recorded as evidence rather than as background colour: what was tried, what happened to it, which readiness dimension it came apart on, and what is different this time. Its scope is the organization's AI history, not only attempts at the use case on this canvas.

Link a prior attempt to a readiness dimension only when the user identifies one unambiguously. If several explanations remain possible, keep that uncertainty. "Nothing has changed yet" leaves the changed-evidence question unanswered; it does not establish that a blocker is resolved.

While a dimension has an attempt recorded against it and nothing recorded about what is different, its readiness status is held at "needs work" and says why. The score is untouched; only what the product says about it changes.

Progressive completion

Progressive completion is a way to reason about evidence maturity, not a universal set of implemented interface statuses. These conceptual stages are useful in review:

• Empty — no usable information yet.
• Seeded — an initial source or answer offers a starting point.
• Partial — some useful context exists, with important gaps.
• Supported — the core content is backed by adequate evidence.
• Decision-ready — the content and remaining uncertainty are sufficient for the next decision.

Build mode counts authored sections with content. Readiness and ROI confidence are separate assessments; the conceptual stages above do not replace those actual interface signals.

Readiness

Readiness measures whether the organization can responsibly move toward implementation. It spans five dimensions: data readiness, process readiness, decision clarity, stakeholder alignment, and governance readiness. Low readiness does not mean the opportunity is bad; it makes the necessary preparation visible.

The 0–100 score shown on the main Brain and AI Readiness pages is the current readiness score. It is not a measure of the Brain's intelligence, a percentage of Canvas completion, or an ROI percentage.

Agentic ROI

Agentic ROI is a Brain-derived, evidence-based assessment of whether the expected value is worth building for and how confident the team can be in that business case. Its 0–100 score measures ROI confidence, not a promised return or ROI percentage.

The assessment considers value potential, baseline evidence, cost complexity, readiness dependency, and payback confidence. It can identify value drivers and missing baselines from existing Canvas evidence, while annual value and payback ranges require baseline inputs that may be supplied measurements or explicitly identified estimates. Evidence remains labeled as validated, estimated, missing a baseline, or unknown.

Agentic ROI is distinct from the authored Business Impact & ROI Canvas section. The section captures the team's impact claims, KPIs, costs, and assumptions; the Agentic ROI view evaluates that evidence together with the rest of the Canvas.

Brain questions

Brain questions are targeted prompts that resolve high-value unknowns. They should be prioritized by how much an answer improves the operating model or unlocks a decision, not by a desire to fill every field.

Memory and connected sources

What I know exposes model memory such as systems, connections, facts and people. Conversation history preserves prior discussions. Connected sources provide external context. A raw source does not automatically become Brain evidence: it must be reviewed and approved as attached memory before its summary and facts can ground answers. Neither memory nor sources remove the need to distinguish sourced facts, inferences, user decisions, and unknowns.

Peer benchmark

A peer benchmark compares a readiness assessment with an anonymized cohort in the same sector. The comparison is directional until the minimum sample size is available. Benchmark records contain sector, project type, readiness scores, and limited outcome metadata—not company names, user ids, or free-text Canvas content.

Projections and artifacts

A projection presents part of the operating model for a purpose or audience. Summaries, slides, workflows, diagrams, blueprints, ROI views, and security assessments are artifacts projected from the same evolving source.

The app organizes them into three hubs: Agentic Brain includes Brain, What I know, AI Readiness, and Agentic ROI; Canvas contains Overview, Summary (what the project is, refreshable), Slides, and Export (every export in one place); Outputs groups the journey into Understand, Prove, Design, and Build, including the Solution blueprint, Solution diagram, Agents, Security Review, and Implementation plan. See Agentic Outputs for generation order and optional execution previews.

Agentic workflow

An agentic workflow describes how AI participates in an operation: triggers, inputs, decisions, actions, systems, human handoffs, exceptions, controls, and outcomes. It is more specific than saying "use an AI agent" and earlier than deploying one.

Security Review

The Security Review is an AI-generated planning assessment of described data categories, potentially applicable regulations, protection strategies, and a high-level checklist. It supports an early governance conversation; it is not legal advice, a compliance certification, a penetration test, or a substitute for review by qualified security and legal teams.

Chat, Guided, and Agent modes

These modes control who approves application within the current Canvas scope. Chat keeps changes behind individual Apply actions. Guided proposes improvements and always asks for approval before applying. Agent executes a visible, bounded sequence and can be stopped. They do not deploy production agents or authorize work outside the Canvas. See Using the Agentic Brain for the surface-specific behavior and approval rules.

For the checks behind these terms, see the agentic AI readiness checklist and how to measure ROI of AI agents. For the sections that fill the model, see the AI use case canvas guide. For what the Brain may and may not change, see governance and safety, and for how evidence enters memory, sources and memory.

Digital twin

A digital twin is the future aspiration: an operating model that remains synchronized with live operational signals and can represent behavior over time. The current product does not claim this capability.

Frequently asked questions

What is the difference between readiness and ROI confidence?

Readiness asks whether the workflow, data, people and controls are prepared for the project to be built responsibly. ROI confidence asks how well the evidence supports the value case. A project can be valuable but unready, or ready but without a compelling value case.

What is an operating model in AI planning?

The operating model is the precise current description of the operation the proposed AI will join: who does what, with which information, under which rules. Canvas builds and edits it, and every output is derived from it.

Return to the documentation index.