I Implemented AI across a 200-person agency as CEO – Here’s what I learned

Key Takeaways

• AI is a powerful assistant, not a decision-maker: Artificial intelligence excels at time-consuming research and data aggregation, saving hundreds of agency hours, but it cannot make high-level strategic decisions.

• The key difference: Pattern-Matching vs. Pattern-Breaking: Generative models are built to replicate existing patterns. Exceptional brand strategy requires breaking those patterns – which remains the exclusive domain of human creativity and insight.

• Responsible innovation: We use AI tools to optimize our internal workflows, but the final shape of creative projects and client relationships is always driven by the empathy and expertise of seasoned professionals.

• The future of the industry: The competitive edge belongs to agencies that learn to seamlessly combine the technological speed of AI with a uniquely human understanding of complex business contexts.

Why did one AI session two years ago change how I lead Admind?

It was July, two years ago. I typed a prompt into ChatGPT – something I had done a few times before, mostly out of curiosity – and asked it to benchmark a competitive landscape for one of our sectors: pulling together what the leading players were doing, how they were positioning themselves, and where the gaps were.

What came back stopped me. Not because it was perfect. It was not. It wasn’t complete either. But in three minutes, AI had done something that in my agency would have occupied a skilled analyst for the better part of a day: it had mapped the competitive field, identified the relevant comparisons, and given me something I could immediately react to and build on.

For a CEO managing an international organisation of over 200 people, time is our most constrained resource. But it was not the time saving that convinced me. It was what the time saving implied: that the entire economics of knowledge work – who does what, at what cost, at what speed, was about to be reorganised. 

From that moment, I was certain: this would be our direction.

What I saw in that first AI-generated report that convinced me

What AI gave me in that first session was a working skeleton – structured, coherent, built in minutes. To understand why that matters, it helps to understand how creative agency work actually flows.

Before any strategic project begins, someone must build the foundation. That means benchmarking: who are the key players, what are they saying, how are they positioned, where does our client sit relative to them, and what has already been tried in this space. In a 200-person agency working across sectors and geographies, this happens dozens of times a month, and each instance absorbs hours of skilled time that delays the moment when genuine thinking can start.

The speed of information gathering and comparison is something most organisations have simply accepted as fixed. 

What I saw in those three minutes was that AI could compress that benchmarking and research phase dramatically moving teams from “gathering and structuring” to “thinking and judging” in a fraction of the time. Anything that accelerates the moment when genuine thinking can begin is not a productivity tool, but a competitive advantage.

The judgment itself stays human

The judgment stays human And understanding why is the part most conversations about AI in creative work skip over.

AI is trained on what already exists. It is, by design, a sophisticated pattern-matching system – extraordinarily good at producing outputs that are statistically consistent with everything it has been trained on. This makes it very good at synthesis, summarisation, and generating options that look like plausible next steps.

It makes it structurally incapable of something else: knowing which direction is right for this client, in this market, at this moment in their history, when the right answer may be precisely the one that breaks with every precedent in the training data.

Brand strategy, at its best, is not pattern-matching. It is pattern-breaking. The insight that repositions a brand, the creative direction that makes a company suddenly visible in a crowded market, the decision to say something no competitor is saying – these require a kind of reasoning that has no analogue in statistical probability. They require a person who has sat with the client’s problems long enough to understand what is not being said, not just what is.

That is the judgment AI cannot replicate. And it is exactly what a branding agency is for.

That is what I mean when I say this is not a productivity tool. A productivity tool makes the same process faster or / and cheaper. What I was looking at could change the quality of the process itself.

AI is a competitive advantage – but only if you understand what it cannot do

There is a version of the AI conversation that treats it as a leveller – a tool that, because everyone can access it, eventually equalises competition. If every agency uses the same models, the argument goes, the advantage disappears.

I think this gets it backwards.

AI does not create competitive advantage by itself. It amplifies whatever capability already exists in the organisation using it. In the hands of a team with weak strategic judgment, AI produces weak strategy faster. In the hands of a team with genuine expertise, it produces better work – because that expertise now has more material to work with, more time to operate, and more room to reach conclusions that would have been impossible under the previous constraints.

This is why the question “will AI replace branding agencies?” misses the point.  AI can tell you what a thousand brands have done before. It cannot tell you what your brand should do next. That gap between precedent and direction is where judgment lives. And judgment is not a soft capability. It is the one thing that compounds over time, that cannot be downloaded, and that no model trained in the past can replicate in the present.

The competitive advantage belongs to organisations that understand this clearly: use AI to eliminate the mechanical work that surrounds judgment, and invest the time recovered into the judgment itself.

That is the model. And it depends entirely on having the right humans in the room.

Where Admind actually started – and what “starting” really means

When I began implementing AI across the agency, our starting point was honest: basic. Some people were already using it quietly, individually, without structure.

The quiet AI adoption problem most agencies don’t talk about

What we were experiencing has a name in organisational research: shadow IT adoption. It is the pattern by which employees integrate tools into their work without organisational sanction, often because the official systems are slower or less capable than what they can access independently. It has existed since the spreadsheet. AI is simply the most recent and most powerful version of it.

We surveyed our team at Admind about how they use AI, what they feel about it, and what is holding them back. The first number stopped me.

96% of our team had already adopted at least one AI tool. ChatGPT alone was being used by 92% of respondents – before we had any formal programme in place. The adoption was already happening. What we were missing was the system around it.

The sentiment picture was more layered than I expected. 57% described themselves as curious and 39% as excited. But 20% expressed concerns about ethics or sustainability, and 17% were neutral. These are thoughtful professionals asking legitimate questions about what we are building and at what cost. That tension is real, and naming it openly matters more than optimising it away.

When we asked what was holding people back from using AI more, the answers were precise. Nearly half (49%)  cited lack of training or knowledge. A further 22% said they were not sure what was allowed. Almost nobody said they did not see the value.

This told us something specific: our people did not need to be convinced of AI’s potential. They already believed in it. What they needed was permission structures, practical training and honest guidelines – not another inspirational talk about the future of work.

The barrier was not scepticism. It was clarity

This is consistent with what large-scale research shows about enterprise AI adoption. McKinsey’s Superagency in the Workplace report (2025) puts a number on the leadership blind spot: C-suite executives estimate that 4% of their employees use generative AI intensively – while employees self-report that figure is actually 13%. Leaders underestimate their teams’ AI adoption by a factor of three. Why? Because without clear guidelines and permission structures, employees experiment quietly, individually, without reporting up. The adoption happens in the shadows – precisely because no one told them whether it was allowed.

What we did in the first 90 days?

Knowing the problem precisely made the solution clearer. If the barrier was clarity (not motivation, not access, not capability) then the response had to be structural, not inspirational.

• First: budget. We invested in tools – not a single company-wide platform that everyone would use the same way, but access to the tools that made sense for different kinds of work. This removed the excuse of cost from the conversation and, importantly, sent a signal: this is something we are serious about, not something we are asking you to figure out on your own time.
• Second: communication. I made our position clear, consistently and repeatedly. AI is not a threat to your role at Admind. It is something we will use, openly, for our benefit and our clients’ benefit. I said this in all-hands meetings. I said it in individual conversations. I said it in the work I showed people I had done myself.
• Third – and this is the part most leadership guides leave out – I showed my own work. Not descriptions of what AI could do. Actual finished work I had built with it: analyses, research structures, presentation frameworks.

There is a psychological reason this matters more than most leaders realise. Behavioral research consistently shows that people adopt new behaviours most durably when they see credible role models using them – not when they are instructed to. When guidance about a new tool comes from above as policy or encouragement, it carries one message: the organisation wants this from you. When the CEO shows their own use cases openly, the message changes: this is something I find genuinely useful in the work I do every day. The first creates compliance. The second creates curiosity. And curiosity is what drives real and lasting adoption.

Budget, communication, leading by example. That was the foundation. But a foundation only matters if what you build on it is designed to last – and what we found next was that the hardest part of implementation had nothing to do with any of those three things.

The hardest part of implementing AI in a creative agency 

Two years in, I can tell you with certainty: the technology is the easy part. Every tool I tested worked as advertised, more or less. Integrations are manageable. Costs are reasonable. The learning curve is steep but not prohibitive.

The hard part – the part that determines whether AI implementation succeeds or quietly fades into the background is human adaptation. And as our own survey confirmed: the top two barriers our team reported had nothing to do with the technology itself.

People don’t resist change. They resist uncertainty.

When AI adoption stalls in a creative team, the instinct is to look at the people. They are too comfortable with the old ways. They do not see the potential. They are protecting their territory.

The question worth asking is not “why aren’t people using AI?” but “what would make using AI the obvious choice?” That shift in framing changes everything about how you respond, because you stop trying to convince people and start trying to create the conditions where the right behaviour becomes the natural one.

At Admind, every time adoption slowed in a particular team, the answer was never more persuasion. It was more specificity: a clearer example, a more concrete use case, a direct conversation about what good looked like in that person’s role. The resistance dissolved not because people changed their minds, but because the ambiguity that was causing the hesitation was removed.

The one thing that works better than any AI training programme

If the goal is creating conditions rather than convincing people, the question becomes: what conditions actually work?

The one thing that consistently moves people is letting them experience AI solving a problem they actually have.

We can call this experiential learning – the principle that we update our mental models most durably when we experience outcomes directly rather than being told about them. The mechanism is simple: a firsthand experience bypasses the rational objections that abstract arguments have to fight through. You cannot talk someone into believing AI will save them three hours. But you can show them, and once they have felt it, the conversation is over.

At Admind, the clearest early example was research synthesis. Brand projects begin with significant groundwork: competitive landscape mapping, sector analysis, client background research. Work that previously absorbed a full day of senior time began taking two to three hours. Nobody needed a presentation explaining why that mattered. They had lived the difference. And people who live a difference tell others about it.

That last part is important. The compounding effect of shared wins – one team discovering something and the rest of the organisation hearing about it is more powerful than any training programme I have designed. Peer-to-peer transfer of working knowledge travels faster and lands more credibly than top-down instruction.

What we still haven’t fully solved

Two years in, AI adoption at Admind is real, and uneven. Some teams have integrated it into how they work. Others are still finding their footing. That is the honest picture, and I think it is worth saying clearly because the public conversation about AI implementation rarely admits it.

The unevenness follows a pattern that researchers of technology adoption have documented for decades. Everett Rogers’ Diffusion of Innovations documented how new ideas spread through populations – the same S-curve pattern, as researchers of organisational change have since observed, appears inside companies too – early adopters move fast, the majority follows when they see proof from people they trust. 

What that means in practice is that the job of a leader is to make the early adopters visible enough that the majority can see the proof they need to move.

At Admind, this is still work in progress. The teams that have gone furthest with AI are not necessarily the ones with the most technical background – they are the ones where a manager created the space to experiment, shared what worked, and made trying feel safe rather than exposed. That pattern, more than any tool or policy, is what we are trying to replicate across the organisation.

How I personally use AI as CEO – and what it’s actually changed

After years of daily AI use, one thing has become clear: how you set up your AI environment matters as much as whether you use it at all.

The tools on my desk (and why I use two different ones)

My primary work environment runs on Claude – one company computer connected to multiple plugins, handling the work that requires integration with systems, data and ongoing processes. Outside of that, I use both Claude and ChatGPT, but for different things. ChatGPT for thinking out loud, exploring ideas conversationally, getting a rough shape of something. Claude for tasks that require holding a large amount of material together with structural precision.

Large language models are trained differently, optimised for different use patterns, and respond differently to different kinds of prompting. ChatGPT tends to be more conversational and generative – good for divergent thinking. Claude tends to handle longer, more complex inputs with greater structural coherence – better for synthesis, holding a large amount of material together.

What I have found, practically, is that having two environments that serve different cognitive modes reduces the friction of getting started on hard thinking. I know which tool to reach for depending on what kind of problem I am facing.

But the tools are only part of the story. The more significant change has been in what AI has done to the quality of my thinking itself – not just the speed of it.

AI as a cognitive multiplier, not a shortcut

This is the distinction that matters most to me, because it fundamentally changes how you think about what AI is for.

There is a difference between automation and augmentation. Automation replaces a human task entirely. Augmentation enhances a human capability – gives it more speed, more range, more precision without removing the human from the process. Most of what AI does in knowledge work is augmentation, not automation. The confusion between the two is the source of both the hype and the anxiety around AI in creative fields.

When I work on a strategic analysis, AI does not replace my judgment about what matters and why. It gives me a richer set of inputs, faster, so that my judgment has more to work with. When I prepare for a complex client conversation, AI does not tell me what to say. It surfaces perspectives I might not have considered, which sharpens what I actually say.

I treat it as a significant acceleration of my cognitive functioning – not a replacement for it. My decisions are always mine. But I believe they are more refined because AI has expanded the range of alternatives I can consider before making them.

The metaphor I find most useful: AI does not make a good chess player irrelevant. It gives a good chess player the ability to consider more moves before deciding. The quality of the decision still depends on the quality of the player.

What implementing AI in a creative agency actually looks like – process by process

The most useful framework we have found is a simple distinction between two kinds of work: high-velocity tasks – work that is rule-based, repeatable, and has a clear measurable output, and high-touch strategic spaces – work that depends on human judgment, accumulated context, and the kind of relational understanding that cannot be specified in a prompt.

The practical test: if you can write down the criteria for a good output in advance, AI can help produce it. If the criteria only become clear once you see the output, and only to someone who deeply understands the client, the judgment stays human.

To give a sense of how this works in practice, here are three areas where the distinction shows up most clearly:

The pattern across all three is the same: AI handles the work that precedes judgment, and humans make the calls that matter. 

This is not a permanent boundary. As AI capabilities develop, the line will shift. The task of a leader is not to defend the current boundary, but to stay clear about where it actually sits and to move resources accordingly as it moves.

The People dimension: what changed in hiring and team expectations

We value openness to AI when we hire. We want people who are curious about these tools, willing to experiment, and honest about what they are learning and not learning.

What has not changed is what we are ultimately hiring for: intelligence, collaborative instinct, professional judgment, and alignment with our values – Cooperation, Trust, Kindness, Partnership, Adventure, Satisfaction.

Our survey showed that even among people who had not yet integrated AI deeply into their work, the vast majority were curious and open. The willingness to engage with something unfamiliar is more valuable than any specific tool proficiency, because tools change. Curiosity compounds.

If I were starting today – one thing I would do differently

I would start sooner. The learning curve is also organisational, not just individual. Teams develop shared mental models about how to use AI well. Those models take time to build, and every month of delay is a month of compounding advantage you are not creating.

Don’t wait for AI to mature

The CEOs I speak to who are still waiting tend to be waiting for AI to mature. I understand the instinct. But consider this: current AI is, in developmental terms, roughly at the stage of an amoeba – in terms of how early we are in the arc of what is coming.

If what already exists – the AI that is transforming how we work – is comparable to an amoeba, then waiting for maturity is not caution. It is a choice to let others build the institutional knowledge, habits and judgment that will determine who benefits most from what comes next.

Start in the right place – ask the right question first

The question worth asking before you buy a single tool or write a single policy is “what does our organisation spend the most cognitive energy on” and where is that energy going to mechanical work that AI could handle?

Our own survey gave us a precise answer. The number one thing our team wanted to hand over to AI was time logging in project management and HR systems. Reports and data processing came second. Emails and routine communication third. 

The first things a creative team wants to automate are the administrative tax on creative work. That is where AI delivers immediate, unambiguous value. Start there.

Before anything else – survey your people

And before you do even that – survey your people. Ask what they are already using, what they are curious about, what concerns them, and what they would most want AI to take off their plate. You will almost certainly discover, as we did, that adoption is already happening without you. The question is whether that experimentation is producing learning that the organisation builds on.

The strategic choice

Two years ago I did not know exactly what implementing AI across Admind would look like. I knew the direction.

What I know now: technology was never the hard part. The hard part was clarity – giving people the permission, the examples, and the honest guidelines that turned curiosity into action. The hard part was patience – accepting that meaningful change moves unevenly. The hard part was staying honest about what AI can and cannot do, and building a practice around that distinction.

What this means for you?

If this article has raised more questions than it answered – that is probably the right outcome. AI implementation is not a project with a finish line. It is a capability you build, unevenly and continuously, over time.

Two places to go from here, depending on where you are:

• If communication is where AI and human judgment intersect most visibly for your team – our free e-book 👉 From AI to Emotion captures what we have learned across 7,000 presentations for clients like ABB, UBS and Accelleron. How to use AI for speed without losing the human impact that makes communication actually work.
• If you are thinking about what AI means for your brand more broadly – how to stay consistent, credible and distinctly human as AI changes how organisations communicate – we are happy to talk. No pitch, no commitment. 👉 Book a conversation with one of our brand consultants about what you are navigating.