Plans

A plan is a step-by-step task flow that tells the AI agent how to help a user complete a guided workflow. "Set up your first workspace," "Configure SSO," "Build a custom report." Each plan is a sequence of tasks the agent executes while narrating what it's doing and why.

What a plan contains

Every plan is a list of tasks. Each task has a type that tells the agent what to do:

  • NAVIGATE: Go to a specific URL or screen in the product
  • CLICK: Click a button, link, menu item, or other interactive element
  • FILL: Enter text into a form field
  • EXPLAIN: Pause and explain something to the user without taking action
  • DECISION: Present the user with a choice and branch based on their answer

A plan for "Create a new survey" might look like: NAVIGATE to the surveys page, CLICK "New Survey," FILL the survey name, EXPLAIN what survey types are available, DECISION for which type to create, then CLICK through the configuration screens.

Each task also carries targeting information (which element on the screen) and a voice script (what the agent says while performing the action).

How plans get created

Plans are generated automatically from capabilities and exploration data. When Floe generates capabilities, it identifies what users can do on your product. Plans take each capability and define the exact steps to execute it.

The generation pipeline:

  1. Exploration maps your product's screens and interactive elements
  2. Capabilities define what users can do
  3. Plans are produced for each capability, specifying the task-by-task execution path

You can view and edit plans in the Plans tab of your site in the dashboard.

Decision branches

Some workflows aren't linear. When a user reaches a point where they need to make a choice, the plan uses a DECISION task.

A DECISION task presents options and branches the plan based on the user's response. For example, if a user is setting up notifications, the plan might ask: "Do you want email notifications, Slack notifications, or both?" Each answer leads to a different set of subsequent tasks.

Branches can converge back to a shared path after the decision-specific steps are done. The agent handles this automatically, picking up the common steps after the branch completes.

This means a single plan can handle multiple paths through a workflow without needing separate plans for each variation.

What the agent does with plans

During a demo or onboarding session, the agent loads the relevant plan and works through it:

  1. The agent reads the current task
  2. If it's NAVIGATE or CLICK, the agent performs the action in the live browser
  3. If it's FILL, the agent enters text (either a pre-set value or something the user provided)
  4. If it's EXPLAIN, the agent pauses to talk through what's on screen
  5. If it's DECISION, the agent asks the user and follows the appropriate branch
  6. After each task, the agent advances to the next one

Throughout this process, the agent narrates what it's doing using conversational AI. It doesn't silently click through screens. It explains the purpose of each step so the user understands the workflow, not just the mechanics.

Template variables

Plans can include placeholder values that get filled in at runtime. If a task says "Fill the project name with {{project_name}}," the agent will ask the user for their project name before executing that step.

This keeps plans reusable. The same "Create a project" plan works for every user because the specific values are collected during the session, enabling personalized demos at scale.

Reviewing plans in the dashboard

The Plans tab shows all generated plans for your site. For each plan you can see:

  • Task list: Every step in order, with type and description
  • Targeting: Which element each CLICK or FILL task targets
  • Voice scripts: What the agent says at each step
  • Decision points: Where branches happen and what the options are
  • Flow view: A visual graph showing the plan's structure, including branches and convergence points

If a plan doesn't look right, check whether the underlying capability and exploration data are accurate. Plans reflect what Floe learned from your product, so improving the inputs improves the plans.

Editing plans

You can edit plans directly in the dashboard:

  • Reorder tasks by dragging them
  • Edit voice scripts to change what the agent says
  • Adjust targeting if the agent is clicking the wrong element
  • Add or remove tasks
  • Modify decision options and branches

Edits take effect immediately. The next demo or onboarding session using that plan will use the updated version.

Plans vs. capabilities

Capabilities describe what users can do. Plans describe how to do it. A capability says "Configure notifications." The plan for that capability says: navigate to Settings, click Notifications, fill in the email field, decide between digest and real-time, and so on.

Every plan is linked to a capability. When the agent decides to show a feature during a demo, it activates the capability's plan and walks through the steps.

FAQ

Are plans required for the demo agent to work? The agent can answer questions and navigate freely without plans, but plans make guided workflows much smoother. Without a plan, the agent improvises each step. With a plan, it follows a tested path.

How many plans does Floe generate? One per capability, roughly. A product with 20 capabilities will have around 20 plans. Complex capabilities might produce plans with 15-20 tasks; simple ones might have 3-5.

Can I create plans manually? Plans are generated from capabilities and exploration data. You can edit generated plans, but creating one from scratch isn't supported yet. The best approach is to ensure your content and exploration cover the workflow, then regenerate.

What happens when my product's UI changes? Re-run exploration, then regenerate capabilities and plans. The new plans will reflect the updated interface. Old plans remain until you regenerate, so there's no disruption to active sessions.