Your agency has AI tools. Now how do you train people to use them?
Neil Perkin, leader of digital consultancy Only Dead Fish, on why upskilling must go beyond understanding to embedding AI into everyday work.
In recent years, AI has presented unavoidable challenges for agencies. The expectation to deliver faster with fewer people has put traditional agency models under strain, while the speed and scale of change have forced agencies to rethink nearly every aspect of how they operate.
Investing in tools and specialist teams is important, but Neil Perkin believes that upskilling staff is a critical step in producing creative and effective work in 2026. As the leader of digital consultancy Only Dead Fish, an AI tutor for several IPA courses and the author of three books on digital transformation, he brings hands-on expertise in helping agencies adapt to this new landscape.
Perkin’s view is clear: AI is not just a tool, it is reshaping the very nature of creative output. Agencies must rethink knowledge management, embed AI fluency across teams, and establish governance structures that balance innovation with human judgment.
We spoke with Perkin to uncover how agencies can empower all practitioners to use AI strategically, creatively and comprehensively - turning adoption into transformation.
How should agencies rethink AI training?
"I think there’s a table stakes here, which is AI fluency. There’s a temptation to just give people the tools and set the expectation that they’ll use them. But there’s a huge difference between using AI tools like a search engine every now and then and having them truly embedded in the way you work."
"It’s about really understanding what AI is good at, what humans are good at and how you combine those strengths in ways that lead to exceptional outputs or outcomes.
Training, therefore, needs to go beyond a basic or fundamental understanding of the tools. It should extend into the areas agencies have always been strong in - different perspectives, creativity, strategy and planning. These are the kinds of disciplines where you can embed AI use cases all the way through the process and into the way people interact and collaborate.
Upskilling is about recognising that AI is a fundamental change that requires people to operate and work in different ways. People need to be supported and trained to do that."
What does effective AI upskilling and embedding look like beyond a one-off workshop?
"One-day workshops are the starting points designed to give people practical use cases, approaches and techniques that they can start applying immediately."
"They can fast-track your understanding of how you might work differently with AI - whether that’s in strategy, planning, creative development or elsewhere. But the real shift happens afterwards. You have to build the habit and apply those techniques consistently in order to truly embed them into the way you work.
The key thing is to take learned techniques and develop an AI habit. It’s about thinking consistently about how you can use those approaches to generate better outputs and better outcomes, and embedding that thinking much more regularly into your workflows."
Does everyone need to be AI-literate or is there still a role for specialist AI teams?
"I think it depends on where you are on the AI implementation journey."
"Specialist individuals or teams can be useful in helping to bring focus, new thinking and momentum into the organisation. They can act as catalysts. But alongside that, everybody is going to have to understand and use AI, simply because of the nature of what it is and what it does. AI fluency is required for everyone, not just specialists."
"The other thing to think about is how you create safe-to-fail environments, and what your guardrails and policies look like. A useful way of framing this is by using the difference between traffic lights and roundabouts:
Traffic lights are your hard and fast rules - red, stop; green, go. They represent clear policies: you can use these tools, you can’t use those; you can do this, but not that. They help people understand exactly where the boundaries are.
Roundabouts, on the other hand, operate more on rules of the road and shared expectations. Here, you’re encouraging people to experiment, to try things out in a safe-to-fail way and to discover how AI can create value in what they do.
It’s not necessarily one size fits all. But everyone should be able to find their own path and their own way of integrating AI seamlessly into their work."
How do you avoid creating a two-tier workforce - AI superusers versus everyone else?
"A lot of this comes down to training and investment in support for people to learn."
"You also need to set the expectation that everyone will embed AI into their workflows and use the tools - but they need the right support to do so effectively.
The risk is setting ambitious expectations about efficiency or productivity gains without actually teaching or supporting people to use the tools in ways that deliver those gains. Without that support, people might see AI as just a shortcut rather than as something that helps them think differently or truly augment what they’re doing."
You’ve written about moving from “knowledge management” to “knowledge architecture.” What does that mean in practical terms for agencies?
"AI is enabling us to connect ideas and people in new and different ways. The risk, though, is that we see AI purely as a route to optimisation and efficiency."
"That’s a natural place to start. But there’s also a huge opportunity in seeing AI as a way to think differently - to open up new creative perspectives and enable us to do things we couldn’t do before.
Knowledge architecture is essentially about how you structure and organise your knowledge in ways that make it genuinely usable. It means making information searchable and accessible, so that anyone in the organisation can find any piece of information that has been created somewhere in the business.
There’s also the knowledge that comes into play creatively. Creativity is combinatorial - it’s about bringing different perspectives together. AI can play a role there too, by helping to organise information, surface unexpected connections and introduce perspectives that perhaps weren’t visible before.
Either way, what AI should be doing is supporting humans behind the wheel of the business."
What skills are at risk in the AI era?
"One of the big risks is what’s sometimes called ‘cognitive outsourcing’ - where people stop thinking and rely solely on an AI engine."
"We’re likely to see a lot of that, because it’s easy to get a “good enough” answer without properly interrogating it or applying any real judgment. You can imagine a document being produced using AI, then passed to someone else who uses AI to summarise it, and so on. Human judgment barely touches the process as they’d be going through the motions without applying critical thinking.
That’s the risk: that thinking itself becomes commodified.
What agencies will need to work hard to protect is human judgment, critical thinking and discernment. The goal shouldn’t be to outsource thinking to AI, but to use AI to augment and amplify human thinking.
That requires a deliberate approach. You have to help people understand how to work with AI in the right way - ethically, thoughtfully and with accountability - so that it enhances capability rather than replacing it."
How do you balance experimentation and innovation with governance, risk and legal responsibility?
"You have to have clear governance and policies in place and they need to come from the centre of the organisation."
"With such fundamental change, people need to understand what they can and can’t do. Once they’re clear on that, they actually feel freer and more confident to experiment and try new things in new ways.
Beyond this, it’s also about setting the right expectations and the right philosophy for how the agency uses AI.
Part of that philosophy might be seeing AI as a way to amplify human capability, skills and creativity - not replace them. It might also mean being realistic about efficiency gains. AI can absolutely drive efficiency, but perhaps not at some of the hyped levels people talk about.
Bias remains a risk as it relates to the datasets models are trained on, so you need to stay aware of what data is shaping the outputs. There have been plenty of studies showing that some major platforms are biased towards Western points of view, for example.
At the same time, we also have to acknowledge human bias. Cognitive biases - like confirmation bias - are equally powerful. We tend to look for information that confirms our existing worldview. With AI, that risk can be amplified, because it’s easy to prompt the system in a way that gives you the answer you were hoping for.
If you’re not careful, that doesn’t lead to the best output or the best decision. So balancing experimentation with governance isn’t just about legal responsibility - it’s about cultivating awareness, judgment and intentional use, both at an organisational level and an individual level."
Are agencies focusing too much on tools and not enough on their operating model and structural change?
"Possibly, yes. There’s naturally a lot of focus on tools. Some agencies have invested in training, which is great, but I think there’s definitely room to go further in terms of capability building."
"In terms of operating model and structure, the changes are so significant that agencies really need to think carefully about how they’re structured, how they operate and what they actually do that provides value to clients - in other words, how they get to their outcomes.
If you think about agents and automation, and how you’re using them to augment capability or deliver outcomes, it’s really about having a clear point of view on what should be automated and what shouldn’t. That clarity will be crucial moving forward."
How should agencies be deploying AI agents?
"Some agency groups have thousands of agents in use, but most of those are fairly straightforward and simple - not particularly sophisticated."
"Where it starts to get really interesting is when you can set an outcome and the agent determines how to achieve it, becoming fully autonomous in that way. That’s when strategic thinking becomes essential: auditing where the real opportunities are within your business and taking a structured approach to agent deployment.
Agencies need to identify where agents can drive proficiency, productivity, creativity, or new perspectives and figure out the balance between agents that everyone uses versus agents created by individuals to support specific tasks."
Looking ahead, what will separate agencies that have genuinely transformed through AI from those that haven’t?
"The agencies that are truly ahead of the curve are the ones making thoughtful investments and really exploring the opportunities AI offers."
"Some agencies are already doing this, using platforms and tools in sophisticated ways. They’re not just chasing efficiencies - they’re thinking differently and creating bigger opportunities for the future.
A key differentiator will be getting the human-AI balance right. These agencies will draw on the strengths that traditional agency people have always excelled at - creativity, perspective, and strategic thinking - and use AI to amplify that.
AI isn’t replacing human skill; it’s enhancing it, helping to add real value to client relationships rather than just delivering things more efficiently."
Built with Shorthand