From Automation to Intelligence: The AI Maturity Ladder

Ask a growing business where it stands with AI and you tend to get one of two answers. Either a list of the tools they have bought, or a slightly guilty version of "we know we should be doing more". Neither answer is much use, because neither is a map. A pile of tools does not tell you how far you have come, and vague guilt does not tell you what to do next.

When I spoke at B2B Remastered this month, a London event for senior B2B marketing and brand leaders, I put up a ladder instead. Seven rungs, from the generative AI everyone has now to the agentic organisation almost nobody has yet. The value of a ladder is that it forces a specific question. Not "are we doing AI?" but "which rung are we standing on, and what does the next one actually need?" This post walks up all seven, and it makes the case that one rung in the middle, the one most businesses climb straight past, is the one that decides whether any of the others hold your weight.

The seven rungs, from doing the work to remembering it

The ladder has a natural break in the middle. The first three rungs are systems that do the work: they produce, they run, they act. The rungs above are systems that remember the work and get better at it over time. Most of the AI conversation, and almost all of the AI spending, sits in the bottom half. The compounding value sits in the top half, and the hinge between them is rung five.

Rung	Capability	What it gives you
1	Generative AI	Content and answers on demand, one person and one chat window at a time
2	Automated workflows	Repeatable tasks that run without you, reliably, in the background
3	AI agents	Multi-step actions taken towards a goal, not just a single output
4	AI platforms	Those agents and workflows connected across the business rather than siloed
5	Intelligence systems	One owned, current, cited memory that everything above it draws on
6	Proactive agents	Systems that act before you ask and learn from what worked and what did not
7	Agentic organisation	The whole business runs this way; people do only the part that needed a person

You do not climb this ladder tidily, one rung at a time, and you will often be standing on several at once. But the sequence matters, because each rung above five quietly depends on five being there.

Rungs one to three: the systems that do the work

The bottom of the ladder is where almost everyone already is, and it is genuinely valuable. Generative AI put a capable assistant in every browser tab, and most teams now use it daily without thinking of it as a project. Automated workflows take the repetitive, cross-system tasks that eat a week, the reporting pulls, the reformatting, the routing, and run them without a person. AI agents go a step further and take multi-step actions towards a goal rather than producing a single output.

These three rungs are what most people mean when they say "we are doing AI", and the agents in particular have arrived faster than most leaders have clocked. Salesforce reports that 83% of organisations now say most or all teams have adopted AI agents, running an average of 12 agents each, a figure it projects to climb 67% within two years. That is a lot of agents doing a lot of work. It is also where the awkward question starts, because none of these rungs answers it.

The question is: what do all those agents actually know about the business they work for?

Rung five is the one everyone skips

Rung four, the AI platform, connects your workflows and agents so they stop being a drawer of disconnected gadgets and start being a system. Useful, and increasingly where the tooling naturally lands. But connection is plumbing, not memory. A platform can route a request between five agents flawlessly and still have every one of them working from a different, partly stale picture of your company.

Which brings us to rung five, the intelligence system: an owned, maintained store of what your business knows, built so that people and AI can ask it questions and get current, cited answers. It is the shared memory the whole ladder above it stands on. We spelled out what an intelligence system is, and how it differs from a wiki or a dashboard, and its absence has a name and a cost: corporate amnesia, the business quietly paying to recreate what it already knew.

Rung five is the one most firms climb straight past, and the reason is almost mechanical. The rungs that do the work are the ones vendors sell, and they deliver a visible win fast. The rung that makes the work trustworthy is nobody's product category. There is no logo for "your business remembers what it knows", so it gets skipped, and the ladder above it is built on air. You cannot have an agent that sounds like you until something, somewhere, holds one current, cited version of what you actually stand for.

What the top of the ladder looks like

Above rung five, the ladder gets genuinely interesting, and this is where the compounding shows up.

Rung six, proactive agents, are systems that act before you ask and, more importantly, learn from the outcome. I run one as a personal obsession: an autonomous agent called Bulldog that wakes itself several times a day, reads the market, argues with itself, and decides. To be honest about it, it trades on paper, not real money, and it is currently behind a simple index fund, which is rather the point. What makes it worth watching is not the trading. It is the library of several hundred cited notes behind it, and what that library does when a trade loses: the evidence gets written back, and the agent quietly changes its own behaviour next time. That only works because there is a memory underneath it. Take rung five away and rung six is just a faster way to repeat your mistakes.

Rung seven, the agentic organisation, is the whole business running this way. It is worth saying clearly what that is not: it is not a company with no people in it. It is a company where the people only do the part that actually needed a person. Agenticise runs close to this itself. We built an intelligence system for our own business, connected and cited, covering our clients, decisions, positioning, and lessons, and it answers a question about the company in roughly 59 milliseconds. What that system actually contains, down to the layers and the folder counts, is worth a look if rung five is where you are standing. This blog is drafted agentically against it, with me reviewing before anything ships. The honest version, the one I gave from the stage, is that no agent runs this company while I sleep. I am above the loop, all day, every day. The automation gets the 90% that never needed me, and the judgement stays human on the 10% that always did. An agentic organisation is a practice, not a purchase, and it took us a year of maintaining that memory to get here.

Why you cannot skip rung five

The temptation, once the agents are cheap and the platform is connected, is to climb fast: more agents, more autonomy, straight for the top. It does not work, and the failure is well documented. Research compiled by marketing technologist Gene De Libero notes that fewer than 1 in 5 companies attempting AI adoption have produced significant tangible impact on their bottom line, and Gartner has predicted that at least 30% of generative AI projects will be abandoned after proof of concept by the end of 2025, with poor data quality and unclear business value among the leading reasons.

Read those two findings together and the diagnosis is not really about models or budgets. It is about the missing rung. An agent is only as good as the context it can reach, and climbing past rung five means asking your agents to act on thin, scattered, or out-of-date information. They do it confidently, and at speed, which is exactly what makes it expensive. Skipping the memory does not save you the step. It just moves the cost downstream and adds a zero.

How to climb from where you are

The ladder is a diagnostic, not a to-do list, and the honest first move is to place yourself on it. Ask where your agents and workflows get their facts. If the answer is "whatever someone pasted into the prompt", you are living on rungs one to three: real systems that do real work, with no shared memory behind them. That is not a failure. It is the most common place for a capable mid-market business to be, and it means the highest-value move available to you is not another agent. It is rung five.

Building the memory is its own piece of work, and we walked through the sequence in the mid-market roadmap for building your own agentic intelligence: document the direction first, then structure the humans around it, then automate the workstreams that cost measurable hours. The ladder tells you where you are standing. That roadmap tells you how to climb. Neither starts with buying more tools, and both start with writing down what your business already knows.

Practical takeaways

1. Place yourself honestly. Which of the seven rungs are you actually standing on? The test is not how many tools you own, it is whether your agents and workflows draw on one shared, current source of truth.
2. Stop counting agents. A dozen agents on a shaky memory is not progress towards rung seven. It is rung three, repeated twelve times, on a wobblier footing.
3. Treat rung five as the priority, not the afterthought. The intelligence system is the highest-value rung precisely because it is the one everyone skips. Building it makes every rung above it possible, and building it so your team owns it outright is the whole case for owning your intelligence.
4. Do not confuse a platform for a memory. Connecting your tools is rung four. Giving them something true and current to work from is rung five. You need both, and only one of them compounds.
5. Climb deliberately. Your AI automation strategy should sequence the rungs, not sprint for the top. The businesses that reach the agentic organisation are the ones that built the memory first.

Frequently asked questions

What is the AI maturity ladder?

It is a way of describing how far a business has actually come with AI, in seven rungs: generative AI, automated workflows, AI agents, AI platforms, intelligence systems, proactive agents, and an agentic organisation. The first three rungs are systems that do the work. The rungs above them are systems that remember it and improve on their own. The point of the ladder is to replace "are we doing AI?" with a more useful question: which rung are we actually standing on, and what does the next one need?

What is rung five, the intelligence system?

Rung five is an owned, maintained store of what your business knows, built so that people and AI can ask it questions and get current, cited answers. It is the memory that everything above it depends on. An AI platform can connect your agents, but it cannot make them sound like your business or act on current facts unless something holds one true, cited version of what you actually stand for. That something is the intelligence system, and it is the rung most firms climb straight past.

Why do most businesses get stuck between automation and agents?

Because the rungs that do the work, workflows and agents, are the ones vendors sell, and they deliver a visible win quickly. The rung that makes them trustworthy, the shared memory, is nobody's product category, so it gets skipped. The result is a business with a dozen agents, each acting on a different and partly stale picture of the company. The stall is not a technology problem. It is a missing foundation, and adding more agents on top of it makes the wobble worse, not better.

Can you skip straight to AI agents or an agentic organisation?

You can buy the tools, but you cannot skip the memory. Every agent you add inherits whatever context your business can give it, and if that context is thin, scattered, or out of date, the agent is confidently wrong at speed. Climbing to proactive agents and an agentic organisation without rung five in place means automating your blind spots. The businesses that get to the top of the ladder are the ones that built the shared, cited source of truth first and pointed the agents at it second.

How do I know which rung my business is on?

Ask where your agents and workflows get their facts. If the answer is "whatever someone pasted into the prompt", you are on rungs one to three: you have systems that do the work but no shared memory behind them. If there is one owned, maintained, cited place your people and tools both draw on, you are on rung five and the rungs above it are open to you. Most mid-market firms are further up on "doing the work" than on "remembering it", and closing that gap is the highest-value move available to them.