Why Plans

Your agent is good at answering questions and handling one-off requests. But when a task has several steps that need to happen in a set order every time, an agent left to its own reasoning will improvise. The same request can produce three steps one day and ten the next. Plans fix that. A Plan is a sequence of steps you define once. When a user’s request matches it, your agent works through those steps in order, gathering what it needs and confirming as it goes. You decide what happens. The agent decides how to carry each step out and adapts to the person it’s talking to.

​

The problem with letting agents improvise

Conversational agents are flexible by design, which is exactly what you want for open-ended questions. For repeatable processes, that same flexibility works against you in three ways.

• Results are inconsistent. The agent might cover five steps for one user and skip two for the next, because nothing holds it to a fixed sequence.
• Your team’s know-how stays trapped. The right way to onboard a new hire, provision access, or resolve a billing query lives in people’s heads or in a document nobody opens. There’s no clean way to put that into the agent’s behavior without fragile prompt engineering.
• When something goes wrong, you can’t tell whether the process was wrong or the execution was wrong, because there was never a defined process to compare against.

​

What a Plan gives you

Plans keep the conversational strengths of your agent and add a reliable backbone underneath.

• Consistency. The steps you define run the same way every time, no matter who triggers the process or how busy the day is.
• Your expertise, encoded. The person who knows the process designs it once. Their best practice becomes something the agent actually does, not just something it’s documented somewhere.
• Structure with room to adapt. The Plan sets out what needs to happen. The agent still interprets each step, asks for missing details, and handles the unusual cases in natural conversation.

​

With and without Plans

​

When to reach for a Plan

Use a Plan when a task has multiple steps that should run the same way every time. Good candidates include:

• Triaging a support request or bug report: gather the details, classify the issue, set the priority, route it to the right team, and confirm back to the person who raised it.
• Onboarding a new employee: create the account, assign the team, set up orientation, send the welcome materials.
• Granting system access by role: check the role, provision the right accounts, set permissions, walk the person through setup.
• Resolving a customer issue: look up the account, identify the problem, check eligibility, process the resolution, confirm.
• Publishing content: review against brand guidelines, check links, generate the social copy, schedule, notify the team.

For one-off questions or tasks where you want the agent to reason freely, you don’t need a Plan. Plans are for the processes you’d want to run identically tomorrow, next month, and after the person who designed them has moved on.

​

Where Skills fit

A Plan defines the steps. A Skill gives the agent knowledge to draw on while it works through a Plan. Skills are reusable knowledge units you attach to a Plan, so the agent has the right framework, guidelines, or reference material to hand while the Plan runs. You can read more in the Skills overview.

​

Next steps