Helping Agents Execute Business Requests
The biggest barrier is usually not “I do not know the command.” It is “I did not describe the business problem clearly enough.” This page helps business users phrase requests in a way that an Agent can act on.
Remember one principle first
You do not need to think about:
- which command should run
- whether a dataset code is needed
- whether this should use SQL or BFF
What you do need to think about is:
- what business problem you are trying to solve
- what final result you want
- whether this request is read-only or allowed to write
Commands, SQL, BFF, and notifications should usually be handled by the Agent and its skills in the background.
One-sentence template
Try to phrase your request like this:
Help me in [which business system or business domain], find or process [which business object], within [time / warehouse / store / project / customer group], under [status or field conditions], and give me [a list / details / summary / recommendation / notification result]. This request is [read-only / allowed to write].
Examples:
- Help me find paid but unshipped orders from the last 7 days, group them by warehouse, and give me an exception list. Read-only.
- Help me find customers created in the last 30 days who still have not placed their first order, group them by industry, and give me a sales follow-up list. Read-only.
- Help me pull recent tickets, orders, and follow-up records by mobile number and turn them into customer-service-ready response notes. Read-only.
- Help me change requirement 241 to
processing. Preview first, then wait for my confirmation before executing.
Five pieces of information that help the most
| Information | What you should try to say |
|---|---|
| Business domain | which system or business scene this belongs to |
| Business object | customers, orders, tickets, requirements, inventory, projects, and so on |
| Scope | time range, warehouse, project, store, customer segment, and so on |
| Result | whether you want a list, summary, suggestion, or notification result |
| Risk | whether this is read-only or allowed to write |
What “not clear enough” looks like
These requests are hard for an Agent to execute well:
Help me check the orders.
See how customers are doing lately.
Help me handle the tickets.
Because they do not say:
- which system
- which time range
- which status or condition
- what final output is expected
- whether writing is allowed
What “clear enough” looks like
These are much easier for an Agent to handle well:
Help me in the store admin system find paid but unshipped orders from the last 7 days, group them by warehouse, and give me an exception list. Read-only.
Help me in CRM find customers created in the last 30 days who still have not placed their first order, group them by industry, and output a sales follow-up list. Read-only.
Help me in the project management system change requirement 241 to high priority. Preview first, then execute.
If you already know these, the Agent can move faster
If you happen to know any of the following, include them:
- order number
- mobile number
- customer ID
- project ID
- business system name
- skill name
But this is only “faster,” not “required.”
If you do not know them, the Agent can still start with discovery.
When you should explicitly say “read-only”
It is a good idea to say “read-only, do not write data” when:
- you are still confirming definitions
- you are investigating something in production
- you want the Agent to explore first
- you are not fully sure the Agent understands the business structure yet
When you should explicitly say “preview first”
It is a good idea to say “preview first, then execute” when:
- doing batch backfills
- doing batch fixes
- changing statuses or ownership
- confirming before deletion
What business users usually do not need to care about
Most of the time, business users do not need to decide in advance:
- whether this should use
data filterorsql exec - which
sqlcodeshould run - which
functionNameshould run - whether the notification should go through Lark or DingTalk
Those choices should usually be made in the background by:
- skills
- the Agent
- the integration or maintenance team
One final suggestion
If you are not sure how to start, just explain the business goal in plain language first.
For example:
I want to know which customers from the last 7 days are worth sales follow-up first.
I want to know why this batch of orders is stuck.
I want customer service to see what recently happened with this customer at a glance.
Starting from the business goal is more important than starting from commands.
Next Steps
- For ready-to-use scenarios: see Common Business Scenarios
- To understand write safety boundaries: see Risk Level Complete Guide