If you have been following this blog - and we choose to believe you have - you will know that last week we spoke about the difference between generative AI - chatbots, prompt-based image generators etc -  and the new wave of 'Agentic AI'.

We thought it would be a good idea to expand on that a little and run through what amounts to five levels of Agentic AI in some detail. If that sounds like jargon, don’t worry - it’s simpler than it sounds.

In plain terms, an Agentic AI is just an AI system that can act more autonomously on your behalf. And like any good team member, these agents come in different levels of capability.

Let’s walk through the five levels - what they are, what they can do, and how real they are today.

Level 1: The Rule Follower

This is the most basic level. Think of it as a glorified script runner. You give it a command or a form to fill in, and it does exactly what it's told. No initiative, no flexibility. It's helpful for automating repetitive tasks like data entry or sending templated responses. Reliable? Yes. But exciting? Not really.

In most SMEs, Level 1 is the most accessible and may even already be in use somewhere in the organisation – even if it's just a simple workflow rule in your CRM or accounting software. It's cost-effective and dependable, but it can’t do anything it hasn’t been explicitly told to do.

Level 2: The Smart Sorter

Here, we add a bit of brains. Level 2 agents can make simple decisions based on patterns they’ve seen before. Think of an email triage bot that recognises when something’s an invoice and routes it to the right folder. It’s still not making plans, but it can classify and sort things on your behalf.

This level is great for basic document handling, sentiment tagging, or flagging potential compliance risks. It augments workflows without fully owning them. For many SMEs, this is a sweet spot – a decent return on investment with relatively low risk.

Level 3: The Task Handler

Here’s where things get interesting. A Level 3 agent can take a high-level instruction like “Create a report on X” and figure out the steps needed to get there. It can search for information, use tools, check results, and adjust its approach as it goes. It’s still working within a defined scope, but it’s capable of goal-directed behaviour.

This is the level where most of the excitement is right now. These agents can handle multi-step knowledge tasks – think due diligence research, financial analysis, or drafting first-pass reports. They won’t replace your team, but they might save them half a day per task.

They’re not perfect. They sometimes get stuck, misinterpret goals, or need a sanity check. But with the right guardrails and oversight, they’re already proving useful in finance, legal, and professional services.

Level 4: The Collaborator

Now we’re into emerging territory. A Level 4 agent isn’t working alone – it’s part of a team. Multiple agents take on different roles and collaborate to solve a problem. One might gather data, another crunch the numbers, another draft a proposal. There’s even a manager agent overseeing the whole process.

The big promise here is scale: what a single agent does slowly, a team of agents can do in parallel. But the orchestration is complex, and errors can cascade if one agent feeds bad info to another. For now, Level 4 is mostly experimental, but it’s getting closer to real-world viability.

For a forward-thinking SME, this might be where you pilot something. Maybe an AI-driven project team that assembles a monthly investor update, or a group of agents that pre-screens deals or candidates before a human review.

Level 5: The Self-Improver

This is the big one. A Level 5 agent can reflect on its performance and improve over time. It can refine its own methods, learn from mistakes, and even invent new tools if it hits a wall. It doesn’t just follow instructions or collaborate; it adapts.

This kind of agent is the stuff of AI think tanks and advanced research labs. It’s the dream of a system that can take on open-ended business goals, figure out what to do, and get better at it the more it tries. It’s also, understandably, the level with the most risk. A self-improving agent needs strong safeguards, because by definition, it’s learning to do things you didn’t explicitly teach it.

For most SMEs, Level 5 is a longer-term horizon. You won’t be delegating your market expansion strategy to an AI just yet – but you might start seeing tools that learn from your team’s behaviour to offer better support next month than they did this month.

So What?

Understanding these levels isn’t just for fun. It helps you cut through the noise when vendors start throwing around terms like "autonomous agent" or "AI co-pilot."

More importantly, it helps you plan. If you’re still wrangling spreadsheets and email threads, Level 1 or 2 might free up hours a week. If you’re drowning in research and first-draft analysis, Level 3 could change your team’s week. And if you want to future-proof your operations, understanding what’s coming at Level 4 and 5 lets you prepare the ground now – with better data hygiene, clearer workflows, and a mindset open to machine collaboration.

AI isn’t just one thing anymore. It’s a ladder. And the climb, step by step, is how smart businesses get ahead.

·- ··

At GiantKelp, we build AI tools which elevate your people and your business. Talk to us to find out how. #GrowLikeKelp

·- ··