Agentic Outputs

The app creates several views from the same evolving Canvas. Their navigation location reflects how they are used; it does not mean they have separate source data.

The three hubs

Agentic Brain

Brain — questions, recommendations, readiness score, history, and bounded actions.
AI Readiness — evidence across data, process, decision, stakeholder, and governance dimensions.
Agentic ROI — value confidence, drivers, baselines, annual value, and payback when supported.

Canvas

Overview — a complete review of the authored Canvas.
Summary — what the project is, what it does, and what is being done about it, above the written overview. It says when the canvas has changed underneath it and can be refreshed.
Slides — a presentation projection.
Export — every download, copy, and print the canvas can produce, in one place.

Agentic Outputs

The hub reads as the journey from idea to a built system, grouped into three stages. Each artifact is one level deep: a card opens its own page, with a back link to the hub.

Understand — what you told us, in one place. Links into the live Canvas views (Summary and Overview). Nothing here is generated, so nothing here goes stale.

Prove — is this worth building, and are you ready? Scores computed live from the Canvas: AI Readiness, Agentic ROI, and the executive Gap Assessment.

Design — the flow, the agents that run it, and a security review of both.

Blueprint — the design itself, in two views one click apart. Diagram is the flow of how agents, people, and systems connect (node shapes distinguish agents, human steps, systems of record, and decisions — a legend decodes them, and edges name the artifact moving along them). Agents is the agent-by-agent specification, told as an ordered story: an overview opens with what the system does end to end, then a numbered rail walks each agent's triggers, inputs and outputs, tools and integrations, guardrails, system instructions, and sample prompts, with handoff links where one agent triggers the next. A Build it panel closes the story and points into Build.
Security Review — a planning assessment of data categories, potentially applicable regulations, protection strategies, and a high-level checklist. It reviews the design rather than sitting in its chain, so it is generated on its own page.

Build — turn the design into something a person can actually use. Two views: Plan, the phased delivery road map with activities, deliverables, and success criteria; and Deploy, which sends the generated agents to an n8n instance you control and gives you a chat interface and run history.

Validate — prove it against cases with known-good answers. Measurement against your own evaluation set, which is what "it works" has to mean.

Hub cards show whether an artifact exists, when it was generated, and a Stale badge when the Canvas changed after generation.

How the flow works

Each stage generates what it owns, and one action per stage generates all of it. Nothing generates because you opened a page — arriving somewhere never spends anything.

flowchart TD
    canvas["The Canvas<br/>authored sections + systems model"]

    subgraph design["Design — the Blueprint page"]
        diagram["Diagram<br/>the flow"]
        agents["Agents<br/>the specification"]
    end

    subgraph build["Build — the Build page"]
        plan["Plan<br/>phased delivery road map"]
        deploy["Deploy<br/>run it on your own n8n"]
    end

    security["Security Review<br/>reviews the design"]
    validate["Validate<br/>cases with known-good answers"]

    canvas -->|"Generate Blueprint"| diagram
    diagram -->|"same action, continues"| agents
    canvas -.->|"also read directly"| agents
    agents -->|"Generate Plan"| plan
    agents -->|"Deploy"| deploy
    agents --> security
    deploy --> validate

Design is one action. "Generate Blueprint" produces the diagram, then the agents. If it is interrupted, the page says which half exists — Diagram ready · Agents pending — and offers Continue, which generates only what is missing rather than starting again.

Build asks for what it needs. The plan is generated from the agents, so Build states the prerequisite and links to the Blueprint when the agents do not exist yet. Deploying needs the same agents.

Each artifact is regenerated on its own. Once something exists, its own refresh control regenerates just that artifact — regenerating the diagram does not discard the agents. Material Canvas changes mark downstream artifacts Stale rather than regenerating them for you.

A failed step says so where it happened, with a Retry, instead of writing its failure into the artifact.

Which page owns what

| Stage | Page | Artifacts | Generated from | |---|---|---|---| | Design | /assets/blueprint | Diagram, Agents | The Canvas | | Design | /assets/security | Security Review | The Canvas and the agents | | Build | /assets/build | Plan, Deploy | The agents | | Validate | /assets/validate | Proof runs | The agents and your evaluation cases |

/assets/workflow is kept as a redirect into the Blueprint's Agents tab, so older links still work.

What the Security Review is not

The Security Review is automated, high-level planning guidance based on what the Canvas describes. It is not legal advice, compliance certification, a threat model, a penetration test, or proof that all applicable regulations have been identified. Qualified security, privacy, legal, and compliance review remains necessary before implementation.

Read-only output conversations

Brain conversations opened from Overview, Summary, Slides, AI Readiness, Agentic ROI, and Agentic Outputs can explain or critique what is present and recommend where to work next. They cannot silently generate or mutate artifacts. Navigate to the owning Builder or output page to make or authorize a change.

Next: Sources and memory.