A good Agent has a clear goal
Start with an outcome.
You should be able to explain clearly:
“This Agent exists to…”
If the objective isn’t clear, the technology probably won’t solve it.
Published 02/11/2026
Author: Rachel Haines - Product Marketing Specialist

AI adoption is moving quickly.
For many organisations, the conversation has already moved beyond“What is Microsoft Copilot?” to something much more practical: Where should we actually be using AI, and when does an Agent make sense?
In our original blog; Copilot vs. Copilot Agents: What is the Difference & Why It Matters, we explored how Microsoft 365 Copilot supports people in their everyday work, while Copilot Agents can provide more specialised support around particular tasks, processes and areas of knowledge.
But understanding the difference is only the beginning.
The next challenge is knowing when to use Copilot, when to build an Agent and when AI might not be the answer at all. At CPS, we believe the best place to start isn’t with the technology.
It’s with the problem.


It’s tempting to start an AI conversation by asking: “Where could we build an Agent?”
But there’s a better question: “Where is work harder than it needs to be?”
Look across your organisation and you’ll probably find processes where people are:
These moments of friction are where the AI conversation should begin because the objective isn’t to introduce more AI.
It’s to make work better.

Microsoft 365 Copilot is designed to work alongside people across familiar Microsoft 365 experiences, helping with everyday tasks such as drafting, summarising, finding information and working with content.
Use Copilot when you want to:
Think of it as: “Help me do this.”
The person is still in control. Copilot is helping them get there faster.
An Agent is more focused.
It can be designed around a particular purpose, task, process, service area or set of information. CPS training material describes Copilot as the generalist, while an Agent is the specialist.
Think of it as: “Help us do this consistently.”
The conversation is moving beyond individual productivity and towards improving how work happens across the organisation.

This is where things get interesting.
Just because something can become an Agent doesn’t necessarily mean it should.
At CPS, we’ve identified some important characteristics that help separate a genuinely useful Agent from automation with an AI label attached to it. Our existing Good Agent framework focuses on context, human involvement, fail-safes and adoption as important design considerations.
Watch our on-demand webinar Beyond The Impossible: What a Good Agent Looks Like
Start with an outcome.
You should be able to explain clearly:
“This Agent exists to…”
If the objective isn’t clear, the technology probably won’t solve it.
The right answer can depend on the user, their role, relevant policies, organisational information and appropriate boundaries.
Context therefore matters.
It’s one of the core characteristics identified within CPS’s Good Agent approach.
AI doesn’t have to mean removing people from the process.
In fact, your Agent may be more valuable when it knows when not to act.
Important decisions may require judgement, approval or escalation.
That’s why human-in-the-loop design is a key part of the CPS approach.
Real-world processes aren’t perfect.
Information can be missing. Requests can be unclear. Circumstances can change.
Good Agent design therefore needs appropriate fail-safes and a way to escalate when the Agent isn’t confident about what should happen next.
Building an Agent is only part of the challenge.
People need to understand:
…adoption should be considered as part of the design, not something added after deployment. CPS’s existing framework identifies adoption as a design consideration.

This might be one of the most important questions to ask.
Not every problem needs an Agent.
If someone simply needs support with everyday productivity, Microsoft 365 Copilot may already be the answer. If the process consists of predictable steps with little need for interpretation, traditional automation may be more appropriate. And if the underlying process isn’t working well in the first place, fixing the process may be more valuable than adding AI to it.
A useful rule of thumb from existing CPS training is to start with Microsoft 365 Copilot for general work and consider an Agent when the task becomes more specialised, repeatable or requires a defined approach.
The question isn’t: “Could we build an Agent?”
It’s: “Would an Agent make this meaningfully better?”
Is this about helping an individual, a team or improving a wider organisational process?
Is it occasional, or is this a repeated challenge?
Don’t build something new if the capability you need already exists in the tools people use today.
A predictable sequence of steps doesn’t automatically require an Agent.
This may indicate that a more specialised Agent experience could add value.
Identify decisions, approvals and exceptions where people still need to take control.
Less administration? Faster responses? Fewer hand-offs? Easier access to information?
Define the outcome before defining the technology.

Supporting people with everyday work.
It’s to rethink where people’s time and expertise are best spent.
The first phase of the Copilot conversation was about understanding what was possible. Now organisations have an opportunity to become much more selective. Instead of asking: “How many Agents could we build?” Start asking: “Which problems are worth solving?”
Then scale what works.
CPS’s existing Good Agent approach similarly recommends starting modularly, proving value and expanding from there.

Copilot and Copilot Agents aren’t either/or, they’re better together.
With CPS, you can build a smarter workplace where:
Let’s take your next step together.
Book an AI discovery session or request a call back today.