Hello again, as promised in my last post, let's try and
delve a bit deeper in each of the components. So, today’s topic is (surprise,
surprise 😎😎😎 !!) Topics
Simply put, a Topic is how I can define the scope, intent, and purpose
of what my agent is allowed to talk about, which tools it should use, and how
to respond within a specific area of expertise.
When I add Topics, I’m really tightening the agent’s brief:
I’m telling the LLM, in its system prompt, “These are the kinds of
conversations you’re here for, and this is how I want you to behave in them."
Good thing is, Topics are reusable in nature and can be used across multiple agents.
Now that we have a little understanding about it, let's dive straight into the system
Navigation: Tools > AI Agent Studio > Topics
It's quiet an extensive and complex (at least for me) topic
to take up as an example. But still a
great example as it has: narrow scope, explicit tool mapping, and clear
parameter handling.
Purpose of this topic is to help an agent decide which functions in the Assignment Information tool to call, and which fields to fetch, based on the user’s question.
There are multiple sets of instructions like –
- It includes “Additional Field Groups” instructions like:
- If the question is about Probation, add ProbationPeriod, ProbationPeriodUnit.
- If it's about Normal Hours, add NormalHours, Frequency.
- It also lists Core Functions the Topic can direct the agent to use:
- There are instructions around fetching data from business objects using functions like -
- getassignmentdetails – basic job, location, department, country, etc.
- getassignmentmanagerdetails – manager info, using AssignmentId.
- getassignmentdetailsbydate – effective-date-driven details using effectiveDate from the user query.
- It also explains how this Topic chains into other Topics/tools, e.g. passing ActionOccurrenceId, EffectiveStartDate, and EffectiveSequence to Assignment Notes, Work Relationship, Collective Agreement, Deep Link Topics, etc.
And at the end very clear guidelines - Do not call
the function with any field that is not in the list above.
============================
There are also much simpler Topics e.g. Assignment Notes if
you want to have a look
Objective of this topic and where it can make sense is when the
user wants employment assignment notes, this Topic tells the agent to use
the ASSIGNMENTNOTES tool with the right parameters
It mandates these inputs:
- pAsgId – Assignment Id.
- pEffectiveStartDate – start date of the specific assignment record.
- pEffSeq – effective sequence.
- pActionOccId – action occurrence id.
- pModId – set to EmploymentInfo.
Notice how the Assignment Information Topic explicitly says
to pass those exact fields on to the Assignment Notes Topic’s tool. This is the
core job of Topics: glue user intent to the right tool calls safely and
deterministically
So again, to summarize, think of this whole AI Agent Studio ecosystem as below
- Agent Team - the end‑to‑end flow you publish (Supervisor, Workflow, etc.)
- Agent - the reasoning “brain” that talks to users and orchestrates tools.
- Topics - the focused “chapters” of knowledge that constrain each agent.
- Tools - business objects, document tools, emails, REST APIs, MCP tools, etc., that fetch or change data.
I guess that’s enough for one post. Stay tuned for the next part of this post where we’ll try and create our very own topic from scratch
Cheers
- how-do-i-use-ai-agent-studio.pdf
- Introducing AI Agent Studio.pdf – July 2025
No comments:
Post a Comment