How it works in 5 steps

Every piece of work in WilsonOS — a book launch, a week of Pinterest posts, a monthly ads review — moves through the same five steps. Learn these once and the rest of the docs are just detail on each.

flowchart LR
  S1["1 · Set up a brand"] --> S2["2 · Talk to your agent"] --> S3["3 · It does the work"] --> S4["4 · You review and publish"] --> S5["5 · It learns, you measure"]
  S5 -. next goal .-> S2

1. Set up a brand — and it gets an agent

   A brand is a workspace: its products, audience, voice, visual rules, customer language, and constraints. Every brand has exactly one agent, and that agent loads the brand’s context at the start of every session. Scout runs Code Pineapple; Prime runs RiseBlox. You describe the brand once; the agent stays in it from then on.

   You decide: what the brand is and who may see it. Go deeper → Brands & agents

2. Talk to your agent

   State a goal in chat — “help more parents discover this book” — and answer the questions it asks. The agent already knows you, the brand, and what is on its plate, so the conversation starts from context, not from zero. When something is genuinely your call, it stops and asks with concrete options rather than guessing.

   You decide: the goal, the constraints, the answers. Go deeper → Conversations

3. It does the work

   The agent picks the flow that fits, tells you which one, and runs its skills: research, writing, design, image and video generation, page building. Anything that costs money waits for your explicit go-ahead. Progress streams back to you as it happens — which flow, which skill, which tool — so a long task is never a black box. Scheduled automations put work through this same step without you asking.

   You decide: whether to spend, and any mid-course correction. Go deeper → Capabilities

4. You review and publish

   Everything the agent makes lands as a versioned deliverable in the Workbench — documents, images, video, pages — with the task, run, and skill that produced it attached. You comment on a specific version, approve a specific version, and publish it through a connection you authorised. The external identifier that comes back is recorded, so results can be matched to exactly what shipped.

   You decide: what is good enough, and what goes out. Go deeper → Work & review

5. It learns, you measure

   Decisions you made in conversation carry forward to the next session. Ratings and business results — visits, sales, clicks, with their source and freshness visible — feed a weekly improvement pass. A skill that produced a strong result is compared against its previous version with evidence before anyone promotes it; nothing rewrites itself because one post did well.

   You decide: which improvements to accept. Go deeper → Memory & learning · Measure

The same five steps, seen from the app

The workspace is organised into five areas that line up with the steps: conversations (step 2), capabilities and automation (step 3), work and deliverables (step 4), business performance and learning (step 5). Brand setup (step 1) lives with the brand itself. Chat sits beside the workspace rather than replacing it; the Workbench is the review experience inside it. Looking at a task, a dashboard, or a file never silently changes what the agent is doing — you can wander without steering.

What stays constant

• You never choose a tool. You state an intent, answer questions, and approve versions. The agent chooses the method and says which one it chose.
• Every “yes” is on a specific thing. A version, a price, a plan — never “the cover” in the abstract.
• Guardrails apply whether or not you are watching. Cost approval, brand rules, and review are the same for a chat request and a scheduled run.
• The loop closes. Results feed the next goal. That is what makes it a system rather than a very good assistant.

Under the hood

Engineers describe this same path as a loop of eight stations on three foundations — trigger, context, capability, execution, artifact, review, learn, measure. It is the honest picture of the machinery, and it lives in Under the hood. You do not need it to use the system.

direction The five steps are the agreed shape of the product. Each linked section carries its own status on what is live today.

Related

Example: create and launch a productThe five steps applied to a product that does not exist yet.

Example: market an existing bookThe five steps applied to marketing a product that does.

Compared to other toolsWhat is different about doing it this way.

The work loopThe eight-station engineering view of the same path.