AI’s come a long way in the last few years.
Not long ago it felt like science fiction. Then it became automation. Then large language models arrived. Now we’re talking about AI agents. I get asked about AI agents a lot. And obviously with FormusPro being a Microsoft partner,
Copilot licensing Copilot agents too.
I get why there’s so many questions at the moment.
One vendor talks about automation. The latest webinar promises full autonomy. Someone else brags about their latest agent. Before long, if you’re not in the thick of it, it must all start to just blur together.
The problem is the phrase ‘
AI’ is being used to describe a lot of different ‘
things’.
Tools get described with the same words. Marketing teams compress distinctions to label everything as ‘AI’. And selling/bragging on social media rewards bold claims.
Meanwhile though, most organisations are just trying to improve a process that’s slow, manual or expensive.
And the thing is, you don’t need a technical deep dive to make sense of this. You just need a way to understand the differing levels of capability.
There’s a big difference between a system that follows rules and one that suggests ideas. There’s an even bigger difference again between something that assists a user and something that can act on its own. And somewhere further along the scale sits the idea of full autonomy, which sounds dramatic but rarely reflects what businesses actually deploy.
So when I get asked about AI agents, the real question usually isn’t “
What is an agent?” It’s closer to “
What should we actually be using?”
That’s what I’m hoping the next few minutes will clarify for you…
I’m not going to drown you in definitions.
Instead, I’ll take a look at the current spectrum of ‘AI’, that runs all the way from straightforward automation through to autonomous systems and explain where tools like Copilot and AI agents sit within it.
Stop Thinking In Buzzwords And Start Thinking In Capability Levels
Rather than treating automation, Copilot, AI agents and full-autonomy as separate trends, to my mind, it’s much more useful to see them as points on the same scale.
They’re not competing ideas and never have been. They’re just different levels of capability.
At one end of the scale, you have systems that follow fixed rules. At the other, systems that can operate with minimal human involvement.
Most of what I see organisations deploying at the moment sits somewhere in the middle. And if you do picture it as a spectrum, it becomes easier to place each term:
- Automation
- AI Assistance
- AI Agents
- Full-autonomy
Each step represents a shift in how much decision-making you’re delegating to software.
That’s the main (only) difference. Not if it’s really “
AI”. Just… how much judgement and action you’re comfortable handing over.
Once you start to frame it that way in your own head, the conversation is easier to change.
You stop asking “
Do we need AI?” and start asking “
What can we do with AI?”
And that’s a far more useful question.
Automation: Clear Rules, Clear Outcomes
Let’s start at the simplest end of the scale.
Automation is nothing more than when a system follows a set of predefined rules. If X happens, do Y.
It’s that simple. There’s no semantic layer. There’s no interpretation. No reasoning or improvisation. If a case gets marked high priority, assign it to a specific queue. If an invoice is overdue by 30 days, trigger a reminder.
Automation is predictable, traceable and controllable.
And yet, in most organisations, it’s still underused, even in 2026.
Automation only works when the process is stable and repeatable. Finance approvals. Case routing. Data synchronisation. Compliance reporting. The kind of tasks that don’t need manual input from your teams… just consistency.
Where it struggles though is when context changes.
If the rules weren’t written to handle an edge case, the system can’t adapt. It fails, or it escalates. With automation, you get reliability in exchange for flexibility, which a lot of the time is exactly what you want.
AI: The System Suggests, But You Decide
Take a step up and we get to what’s we’ve all been referring to as AI since ChatGPT launched – Large Language Models.
It’s where tools like
Microsoft Copilot, ChatGPT, Claude, Gemini and all the others sit.
The main difference between automation, and the new LLMs is that they don’t just follow pre-determined rules.
They have semantic layers (some much better than others) that can interpret context and generate suggestions. They can draft emails, summarise meetings, recommend next steps, analyse trends and even surface risks.
But, and for me this is the part that matters, they won’t ever act independently.
You’re still in the loop. The LLM will suggest something. But you need to review it (please always do this). You still need to decide what to do with it.
That’s still a huge shift over from automation though.
The system is no longer executing predefined instructions. It can help you think. It’s augmenting your judgement.
Inside platforms like
Microsoft Dynamics 365, with built in Copilot, that might mean:
- Drafting a response to a customer enquiry
- Suggesting follow-up actions based on deal history
- Highlighting anomalies in financial data
- Summarising long case notes into something usable
The underlying process hasn’t disappeared. The human hasn’t disappeared. The speed and surface area of support has simply expanded. That’s why AI assistance tends to be the most comfortable entry point for most organisations. It increases productivity without removing control.
Where it falls short is ‘active action’.
An LLM like Copilot won’t autonomously update multiple systems. It won’t pursue an objective without prompting. It won’t decide that a workflow needs changing and implement it.
It waits for your instructions.
And that’s the line between LLM’s and Agents.
AI Agents: When The System Can Decide And Act
This is where things start to get interesting.
An AI agent won’t just suggest. It won’t wait for you to approve every step. Once you set it going, it can interpret a goal, decide what needs to happen, and take action across multiple systems.
That’s huge!
So instead of: “
Here’s a draft” that you edit and send, it becomes, “
I understand your objective. I’ll handle it and let you know if something needs escalation.”
That’s a broad remit, so in practical terms, that might look like…
An agent monitoring overdue invoices. It checks payment history, reviews prior communications, drafts a reminder, updates the
CRM record, flags high-risk accounts and escalates anything unusual to a human.
No single rule triggered it. No one clicked a button, or approved steps between those actions. It worked towards your specified outcome.
Another good example might be an internal service desk agent.
It will read incoming requests, categorise them, check existing knowledge articles, resolves common issues automatically, books appointments where needed and only pass on the complex cases to a person.
That’s not automation anymore as automation needs a script to follow. An AI Agent doesn’t. It evaluates context, chooses actions and then adapts as it goes.
And that’s why they’re so in demand at the moment, and why I’m encouraging all our clients to reach out and speak to us about them.
They massively reduce friction across multiple systems at once. They don’t live in a single workflow. They operate across data, processes and applications and save organisations a fortune in time and staffing.
Combined with the Microsoft ecosystem… Dynamics 365 and the
Power Platform being a great example, that opens up serious possibilities.
Agents can:
- Orchestrate multi-step processes
- Coordinate between CRM, finance and service systems
- React to real-world changes
- Escalate intelligently rather than blindly
But here’s the bit to remember…
They’re still not weird, sci-fi like rogue intelligences. They still operate within guardrails. Guardrails defined by you. You have to clearly set their boundaries and escalation points as an Agent doesn’t remove the need for governance. It just changes the point at where human involvement happens.
Instead of being involved in every step, your people can focus on exceptions, oversight and optimisation, which let’s face it, is a far better use of time.
And for organisations looking to scale without endlessly increasing headcount, that’s where the real value starts to appear.
Full Autonomy: It Sounds Bigger Than It Usually Is
When people hear the word “
agent”, they often leap straight to full-autonomy.
Systems running entire departments. AI setting its own objectives. Software making strategic decisions without oversight.
That’s not an agent… it’s full-autonomy.
In theory, it means a system that can define goals, optimise towards them and then also operate without continuous human involvement.
In practice, very few organisations are anywhere near that. And most shouldn’t be.
Businesses operate within regulation, financial controls, brand risk and accountability structures. Decisions have consequences. Customers expect traceability. Boards expect governance. Full autonomy removes a level of human oversight that many organisations aren’t comfortable giving up.
And that’s right. Commonplace full-autonomy is a good few years away yet.
I’m mentioning it here though as I’ve noticed a tendency to frame autonomy as the end goal. As if anything less is a stepping stone. But for most real-world processes, autonomy isn’t and shouldn’t be the objective. Reliability, efficiency and intelligent support are. And that’s where AI agents shine.
FormusPro want to help you design systems that can act independently … but within defined boundaries. Systems that will still require approval for high-risk actions. Systems with full audit trails.
That’s where the real ROI is at the moment.
So What Should You Be Exploring Right Now?
If you step back from the fads and trends, most organisations don’t need a full “
AI transformation”.
Some processes will be perfect for automation. They’ll be repetitive, rules-based and stable. Get those right first. That’s your low hanging fruit.
Many teams will benefit immediately from AI assistance. Drafting, summarising, analysing, surfacing insight. That’s where tools like Microsoft Copilot tend to create quick, visible wins. Feel free to reach out to me and we can discuss how best to arrange this at scale for you.
But the real opportunity emerging now is in AI agents.
Not full autonomy. Not science fiction.
Targeted, outcome-driven agents operating within clear boundaries.
The kind that can:
- Manage parts of a revenue cycle
- Orchestrate service workflows across systems
- Monitor risk and escalate intelligently
- Coordinate data between departments without constant human prompting
The question isn’t whether agents should exist in your organisation. It’s where they’ll make the most sense. Which processes are currently manual but structured? Where are your teams overwhelmed by coordination rather than creativity? Where does decision-making follow patterns that could be delegated safely?
That’s our starting point. That’s where you should focus over the next couple of years.
