If a client can open ChatGPT on Monday morning, ask it for 30 campaign territories, generate the visuals, rewrite the copy in half a dozen tones of voice, translate everything, produce variations for every social format and then create a surprisingly convincing presentation explaining the strategic thinking behind the whole thing, what exactly is the agency invoice for?
A few years ago, there were reassuring answers available. AI imagery looked weird. Copy sounded synthetic. Video barely held together for more than a few seconds. Anyone who’d spent their career in a creative department could point at the output, smile indulgently and explain why the machines were nowhere near replacing professionals. That argument has aged rather quickly. The technology is still inconsistent, sometimes spectacularly wrong and frequently more impressive at first glance than it is under proper scrutiny, but it has also become genuinely useful. More importantly, it’s become useful to clients, not merely agencies.
Generative AI is now embedded across marketing rather than sitting outside it waiting to disrupt things. 99% of advertisers surveyed by the ISBA earlier this year were engaging with generative AI in some form, with 65% of individuals using it regularly. In the US agency market, Forrester found nine in ten agencies were already using generative AI, while half were using agentic AI for marketing execution. The question, therefore, is no longer whether advertising and marketing will “adopt AI”. That bit has happened. The more interesting question is what happens once both sides of the client-agency relationship have access to many of the same capabilities.
That changes the argument considerably. A creative agency can no longer answer concerns about AI by proudly announcing that it has AI too. So does the client. Increasingly, so does the client’s media platform, CRM provider, ecommerce system, analytics software and probably whichever application they used to take the meeting notes. AI literacy will matter enormously, but it’s rapidly becoming something closer to digital literacy than a distinctive agency proposition. You need it. You’ll look alarmingly behind the curve without it. But “we know how to use AI” is unlikely to remain a particularly compelling answer to the question of why a company should pay you hundreds of thousands of pounds.
Perhaps that’s why the apparently simple SEO questions appearing around this subject are actually quite revealing. AI or marketing agency? AI or advertising agency? AI or creative agency? They sound like binary choices, but they’re really questions about which parts of the old agency proposition still justify an external partner now that access to production has become dramatically cheaper.
The uncomfortable answer is that some of those parts probably don’t and agencies might be much better off admitting it.
What Has AI Actually Changed About Agency Work?
Jairo Grinberg
The most obvious thing AI has changed is production. An extraordinary range of activities that once required specialist time, specialist software or specialist people can now be accelerated, partially automated or at least started by someone with a browser window and a reasonably clear instruction. Copy drafts, rough concepts, research summaries, storyboards, translations, social variations, moodboards, synthetic photography, presentations and increasingly sophisticated video can all be generated at speeds that would have sounded ridiculous even five years ago.
That doesn’t mean expertise has disappeared from those activities. Anyone who has spent enough time with generative systems will know how quickly the illusion of effortless competence falls apart once the task becomes specific. The fourth version of an image inexplicably loses the product. A paragraph that sounds authoritative contains a fact nobody can source. Brand language becomes slowly homogenised into the same frictionless corporate mush. The impressive five-minute prototype turns into a three-hour wrestling match with consistency. AI removes work, but it also creates different work.
Nevertheless, it has altered the economics. Until relatively recently, one very practical reason to hire an agency was that an agency had access to capabilities the client didn’t have internally. It had copywriters, designers, strategists, planners, producers, developers, media specialists, researchers and networks of partners capable of turning a vague ambition into something that could actually appear in the world. There was value in the thinking, obviously, but there was also enormous value in simply having enough skilled people available to make things happen.
That production capacity is no longer as scarce as it once was. AI is commoditising a certain kind of competent creative output. It isn’t necessarily replacing the best work. It’s doing something potentially more disruptive to agency economics by making the acceptable middle dramatically easier to produce. A perfectly respectable social visual, product description, presentation image or collection of headline variations may simply not be worth what it was when producing it required significantly more time and labour.
Forrester’s 2026 research found that productivity remained the dominant motivation for agency AI adoption, while warning that an excessive focus on cost reduction risks sacrificing the creativity and differentiation that brands ultimately need. That distinction matters because AI can make an agency faster while simultaneously making its proposition less interesting. If every agency adopts broadly comparable tools to shave broadly comparable amounts of time from broadly comparable processes, efficiency quickly stops looking like differentiation and starts looking like the new cost of admission.
The deeper change, then, isn’t that AI has suddenly learned how to “be creative”. It’s that the labour involved in producing many creative outputs is becoming less tightly connected to the price a client can reasonably be expected to pay for them. That causes obvious problems for a business model historically built around people, hours and deliverables, but it also has a potentially healthy consequence. It forces agencies to unbundle two things that have spent decades being sold together: the value of making something and the value of knowing what ought to be made.
Those two things were never identical.
AI has simply made the gap between them much harder to ignore.
What Can a Client Now Do Without an Agency?
vStream Digital Media
Quite a lot, if we’re being sensible about it.
There’s very little value in agencies pretending otherwise because clients are already finding out for themselves. An experienced in-house marketing team with a clear brand, good data and established processes can now produce and adapt work at a level that would have required far more external support only a few years ago. A smaller company can generate competent social content without putting a designer on retainer. Ecommerce businesses can accelerate product copy and imagery. Internal teams can produce rough visualisations, interrogate documents, adapt existing campaign material, develop presentations and explore initial concepts without every task becoming an agency brief.
None of this means those outputs will automatically be excellent. It means they may be perfectly adequate for the job in front of them, and “adequate for the job” is a category agencies sometimes struggle to acknowledge because there’s no particularly elegant way of saying that the client doesn’t need award-winning creative direction to announce revised Easter opening hours.
There are plenty of marketing tasks where external agency involvement is becoming an unnecessary overhead. If the business already knows what it wants to communicate, understands the audience, has clear brand parameters and simply needs twenty variations of something that already exists, AI combined with a capable internal team may genuinely be the better answer. That isn’t the collapse of creativity. It’s automation doing exactly what automation is supposed to do: removing cost and friction from routine activity.
The interesting complication is that widespread use hasn’t yet translated into equally widespread impact. ISBA’s latest research found that while almost every advertiser surveyed was engaging with generative AI, only 28% of individual respondents said it had meaningfully changed their day-to-day work and just 14% of advertisers reported a significant impact on business results. Organisations primarily focused on effectiveness rather than efficiency were substantially more likely to report meaningful benefits.
That gap between adoption and impact should be familiar to anyone who lived through the earlier years of digital transformation, when companies spent enormous sums acquiring technology only to discover that buying the software was the easy part. AI gives organisations extraordinary capability, but capability still has to be integrated with strategy, data, processes, governance, measurement and people. Generating 200 pieces of content isn’t much of a commercial breakthrough if nobody has established whether the company needed them, whether anybody wants to see them or whether they’re reinforcing anything distinctive about the brand.
This is why asking “Can AI replace a marketing agency?” is probably the wrong question. AI can replace agency tasks. It can reduce the need for certain agency services. It can allow internal teams to do things that previously required external help, and it will almost certainly continue eating into executional work as the systems become more capable. Whether that amounts to replacing an agency depends entirely on what the agency believes it exists to do.
If the agency exists mainly to receive instructions and turn them into assets, the outlook is uncomfortable. If its role begins before the instructions have been decided and continues after the assets have gone live, there’s rather more to discuss.
Making the Work Is Not the Same as Knowing What Work to Make
Zara Picken
For most of advertising history, creative scarcity was partly a production problem. Ideas cost money to realise, which meant choices had to be made relatively early. You couldn’t sensibly produce hundreds of fully formed campaign routes simply to see which one everybody preferred. The expense and time involved in execution forced people to narrow the field long before the work looked finished.
Generative AI turns that relationship on its head. A half-formed thought can now arrive looking alarmingly complete. Instead of three rough territories on a wall, you can have 50 immaculate images with cinematic lighting, beautifully presented copy and enough surface polish to make every idea look as though somebody has already signed it off. Suddenly the scarce resource isn’t the ability to produce options. It’s the ability to reject them intelligently.
That sounds like a minor distinction, but it could become one of the most important changes in creative work. When output is expensive, producing something demonstrates a degree of commitment. When output is close to infinite, production tells you almost nothing. The folder containing 400 campaign ideas might represent a brilliantly expansive creative process or simply prove that somebody left a model running while they went to lunch.
The difficult part moves upstream into judgement. Which of those ideas reflects something true about the audience? Which genuinely belongs to this brand rather than any vaguely similar company? Which feels familiar because it taps into an important cultural code and which feels familiar because the training data has quietly driven everybody towards the same answer? Which deserves investment? Which looks impressive as a generated image but collapses the second you try to build a proper campaign around it?
This is where the word “creativity” can become slightly unhelpful because it encourages the industry to defend human value in almost mystical terms. We don’t need to insist that machines are constitutionally incapable of producing anything creative. They plainly can produce outputs people find surprising, interesting or useful, and they’re getting better at it. The stronger case for human creative expertise lies in context and discrimination: understanding why one thing is worth pursuing when hundreds of alternatives are available.
We’re already seeing evidence of the tension. Forrester has warned that an excessive focus on AI-driven productivity can undermine differentiation, while WFA research has similarly captured marketer concerns about the possibility of growing creative sameness. The problem isn’t that generative systems can only make bad work. It’s almost the opposite. They’re exceptionally capable of generating work that feels plausible, familiar and professionally resolved, which can make mediocrity considerably harder to spot.
A useful agency therefore shouldn’t simply promise the client more ideas. The client can have more ideas than it could possibly use before breakfast. The agency’s value lies in understanding the business and cultural context well enough to establish what constitutes a good idea in the first place. That means knowing the category without simply reproducing its conventions, understanding the brand without becoming trapped by its history, spotting what competitors are doing and recognising when the entire competitive set has collectively disappeared down the same strategic cul-de-sac.
It also means being prepared to question the premise of the brief.
This might be one of the clearest areas where an external partner still earns its keep. AI is extremely good at helping somebody answer the question they ask. A capable agency should occasionally tell them they’ve asked the wrong question.
If a brand briefs an AI system to develop a campaign demonstrating its authenticity to Gen Z, it will diligently get to work. It may research attitudes, propose territories, generate language and visualise executions. A good strategist is more likely to spend some time wondering why the brand has suddenly decided “authenticity” is the answer, whether anyone in Gen Z describes the category that way, whether the audience definition is meaningful and whether there’s a commercial problem underneath the marketing problem that nobody particularly wants to discuss.
That’s not merely creativity. It’s diagnosis.
And diagnosis remains considerably more valuable than producing treatment for the wrong condition.
So What Is a Client Actually Paying an Agency For?
Leandra Meintjes
This is where the latest client research becomes genuinely interesting, because brands themselves don’t appear to believe that agencies are about to disappear.
The IPA surveyed 200 senior UK brand-side decision-makers in 2026, alongside qualitative interviews with senior marketers, agency leaders and other industry figures. Nine in ten respondents said agencies were either central to their commercial success or made a meaningful contribution to it, while a similar proportion expected their use of agencies to remain stable or increase over the following three years. Asset production still mattered enormously, as did specialist channel expertise, but strategic brand expertise sat very close to the top of the list.
The more revealing part of the research, though, was the gap between what clients said they valued and what they believed agencies actually delivered. Clients placed considerable value on agencies stretching their thinking and challenging assumptions, yet far fewer felt they were consistently receiving that challenge. In other words, clients aren’t necessarily saying, “We don’t need agencies now because we have AI.” They’re saying something potentially more uncomfortable: “If we’re still going to pay for agencies, we need more from them than execution.”
That should be encouraging and terrifying in roughly equal measure.
Encouraging, because there is clearly still demand for an external creative and strategic partner. Terrifying, because the value being described is much harder to manufacture than another hundred social assets. It requires senior people with judgement, category knowledge, curiosity, confidence and enough independence to disagree with the person paying the bill when disagreement is useful.
There’s an advantage in being outside the organisation that agencies sometimes underestimate because they’re so busy demonstrating how deeply they understand the client. Internal teams will always understand certain things better. They live with the politics, processes, customers, systems and commercial realities every day. An agency can never entirely reproduce that knowledge, and pretending otherwise usually leads to embarrassing moments in workshops.
What the agency can provide is distance.
A good external creative partner gets close enough to understand the problem but remains far enough away to see the assumptions surrounding it. It can recognise that something the company treats as an immutable truth is actually just the way things have always been done. It can introduce ideas from other categories. It can spot when a proposition makes complete sense internally and absolutely none outside the building. It can tell the client that the reason nobody understands the campaign might not be that consumers need “educating”, but that the campaign is confusing.
That kind of productive challenge has always been part of the best agency relationships, but AI makes it more economically important because execution itself is becoming easier. When clients could justify an agency partly on the basis of production capacity, strategic value could occasionally remain rather fuzzy. As that production advantage narrows, the thinking has nowhere left to hide.
Perhaps that’s a good thing.
It means the question “What does an agency offer that AI cannot?” doesn’t need to be answered with some grand claim about uniquely human souls or irreplaceable creativity. The answer can be much more practical. A good agency offers a system of informed judgement and responsibility around the work. It connects business context, customer understanding, cultural knowledge, creative expertise, media reality and implementation. It makes recommendations and can explain why it made them. It carries some accountability for the consequences.
The deliverable might still be an advert.
What the client is actually buying is the thinking that makes one advert worth producing instead of the other 99.
When Is AI Enough, and When Does a Client Need an Agency?
Think Design
The agency industry would probably become more credible if it became much more comfortable admitting that sometimes AI is enough.
Not every business requires a major agency relationship. Not every project needs a strategic reset. Not every Instagram post deserves a discovery phase. If the marketing challenge is clear, the brand is established, the internal team is capable and the work is primarily executional, bringing in an external agency can genuinely add cost without adding much value. AI will make those cases increasingly common.
This is likely to change the shape of agency relationships rather than simply destroy them. Marketers continue to use agencies extensively even as AI enables more activity to be brought in-house. The pattern is one of increasing selectivity: external partners remain important, but marketers are becoming more deliberate about where they spend agency money.
That feels like a much more plausible future than either extreme usually offered in the AI debate. Clients won’t outsource everything simply because agencies have better tools, nor will they suddenly internalise every aspect of marketing because somebody bought an enterprise AI licence. They’ll increasingly decide task by task, capability by capability and problem by problem where external expertise is worth paying for.
A fairly useful dividing line is the cost of being wrong.
If a team needs another version of a display advert built from an existing campaign system, the cost of a mediocre first attempt is low. Generate it, review it, fix it and move on. If a company is repositioning a major brand, entering an unfamiliar market, launching a category-changing product or trying to recover from declining relevance, the consequences of making the wrong strategic decision are rather larger. Suddenly, the speed with which the first execution can be produced isn’t the most important variable.
Complexity matters too. The closer marketing moves towards orchestration across multiple audiences, channels, markets, datasets and organisational stakeholders, the less useful it becomes to think of AI as a magic box that simply “does the marketing”. Forrester’s more recent analysis of AI agents makes precisely this point: agentic capabilities are becoming pervasive, but organisations still need to implement, manage, monitor and connect them sensibly rather than assuming an infinite collection of automated workers somehow organises itself.
The agency opportunity may therefore move away from being the place that physically produces everything and towards being the place that helps make the system coherent. That can include creative leadership, strategic direction, specialist expertise, cultural understanding, integration across partners and the ability to maintain a recognisable brand while the machinery beneath marketing becomes dramatically more automated.
This is also why the threat from AI is not distributed evenly across agency work. The closer the service is to repeatable execution with clear inputs and outputs, the more exposed it is. The closer it gets to ambiguous problems involving multiple competing considerations, the more valuable experienced human judgement becomes.
That doesn’t mean ambiguity is somehow immune to AI. Far from it. Agencies will use AI extensively inside that work too. The important difference is that the technology becomes part of the capability rather than the entire proposition.
Agencies Have to Prove Their Value Differently
Leagas Delaney
This is probably where the conversation gets hardest because it eventually reaches money.
For decades, agency economics have been tied in one way or another to labour. Even where the client wasn’t literally buying an hourly rate, staffing levels, utilisation and time were fundamental to how work was scoped, priced and delivered. AI puts obvious pressure on that arrangement. If something that used to occupy three people for two days can now be completed by one person in an afternoon, the agency cannot simultaneously boast about its 80% efficiency improvement and expect nobody in procurement to wonder why the bill looks exactly the same.
There are only so many times you can explain that the client is paying for “value, not hours” before somebody reasonably asks you to demonstrate the value.
That isn’t an argument for a race to the bottom. Quite the opposite. It’s an argument for becoming far more precise about what the client is actually paying for. Routine production should become cheaper where technology has genuinely reduced the resources required to produce it. Agencies shouldn’t be embarrassed by that. The industry has spent years telling clients that innovation creates efficiency, so it would be slightly cheeky to discover that efficiency is only desirable when the agency gets to keep all of it.
But reducing the price of commoditised work doesn’t necessarily mean reducing the value of everything surrounding it. Strategic expertise, specialist knowledge, distinctive creative thinking, intellectual property, senior judgement and the ability to manage difficult decisions may become more important precisely because the mechanical work surrounding them has become cheaper.
That requires agencies to stop making AI efficiency itself the centrepiece of their proposition. Every case study cannot end with “and we produced 64% more content in half the time”. Sometimes that will be genuinely useful, but it’s still an operational benefit, not necessarily a marketing outcome. Forrester’s research is increasingly explicit on this point, arguing that agencies and CMOs risk confusing productivity gains with effectiveness if they focus too heavily on reducing costs rather than improving differentiation, performance and growth.
The same lesson appears in the advertiser data. ISBA’s findings suggest the organisations getting more meaningful impact from generative AI are those concentrating on effectiveness rather than treating efficiency as the end goal. The distinction sounds almost painfully obvious when written down, but it cuts directly against the way AI has frequently been sold during the past few years. Faster isn’t a strategy. Cheaper isn’t an idea. Producing ten times as much work is only valuable if somebody has established a reason for the extra nine-tenths to exist.
For agencies, this means the pitch has to change. Instead of showing clients an elaborate proprietary workflow full of glowing boxes and arrows, they need to demonstrate what that workflow allows them to understand, decide or achieve that would otherwise have been harder. If AI helps a strategy team interrogate thousands of customer reviews, the interesting thing isn’t the size of the dataset. It’s the insight the team found and what that insight changed. If generative tools allow a creative department to explore 100 territories, the value doesn’t lie in proudly showing all 100. It lies in the quality of the route that survived.
This is also why every agency racing to build its own branded AI interface risks creating a strange new form of sameness. In many ways, AI is actually making agencies less interesting, because once everyone has a proprietary-sounding platform promising to “accelerate”, “augment” and “orchestrate”, the technology stops telling clients very much about the agency behind it.
The tool is not the personality.
The quality of the decisions made with it might be.
So Why Should a Client Still Hire Your Agency?
D8 Amsterdam
The most useful answer to this question probably begins with a concession: if all a client needs from your agency is the physical production of the work, increasingly there may be no particularly strong reason for them to hire you.
That doesn’t mean production has no value, nor that designers, writers, directors, animators, developers and all the other people involved in making things are suddenly irrelevant. It means the economics around a significant proportion of routine output are changing quickly. Clients will rightly expect technology to remove labour where labour no longer contributes proportionate value, and agencies that attempt to preserve every historic revenue stream by pretending otherwise are likely to find the conversation rather difficult.
The opportunity is to recognise that making the work was never the most interesting part of the best agency relationships anyway. The agency mattered because it helped a client see something differently. It understood the company well enough to know what mattered but remained sufficiently outside it to ask inconvenient questions. It brought people together who thought about the problem from different angles. It translated messy commercial realities into a strategy and then translated that strategy into something an audience might actually care about.
None of that disappears because an image can now be generated in thirty seconds.
If anything, the explosion of production capability makes coherence more important. The average marketing organisation is moving towards a world in which it can create far more than it could previously afford to create. It can personalise, localise, test, remix and optimise at extraordinary scale. It can ask models for endless interpretations of the same brief and deploy agents across activities that once required teams of people. The natural consequence is not simply “more creativity”. It is also more noise, more choices, more complexity and more opportunities for a brand to slowly dissolve into an enormous pile of individually competent but collectively meaningless content.
That may be where agency value moves next.
The agency doesn’t have to be the factory producing every item. It can become more valuable as the intelligence helping the client decide which items should exist, how they relate to one another and what they are collectively trying to accomplish. It can help an organisation resist the understandable temptation to confuse abundance with effectiveness and personalisation with relevance. It can make sure a brand still has an identifiable point of view after 40 automated systems have been given permission to speak on its behalf.
This isn’t a convenient fantasy invented by agencies to protect themselves. The current client research suggests brands still see meaningful commercial value in agency partnerships while simultaneously asking those partners for more challenge, stronger strategic thinking and a clearer contribution beyond execution. The relationship is not disappearing so much as being renegotiated.
Agencies therefore shouldn’t spend the next few years trying to prove that clients are incapable of using AI themselves. They plainly are capable, and the tools will only become easier. Nor should the industry retreat into a comforting argument that human creativity possesses some magical quality machines can never touch. That might be philosophically interesting, but it’s a fairly weak commercial proposition when a client has just generated something they quite like for £20.
The stronger case is much more grounded. Agencies can be valuable because businesses don’t merely need outputs. They need choices to be made under uncertainty. They need somebody to connect cultural signals with commercial realities, to spot assumptions, interrogate research, understand trade-offs, bring specialist expertise into the room and take responsibility for recommending one course of action over another. They need people who are capable of using AI extensively without automatically accepting whatever it produces, and who understand enough about brands, audiences and creativity to recognise when the easiest answer is taking everybody in the wrong direction.
That changes what a good agency looks like. It probably becomes leaner in some places and more senior in others. Certain production revenues will shrink while strategic and specialist capabilities become more important. Some client relationships will get smaller because the client can sensibly do more itself. Others may become deeper because the complexity of orchestrating increasingly automated marketing creates problems that no single internal team wants to solve alone.
There is no guarantee that every agency makes that transition. Some were built around precisely the kind of execution AI is commoditising, and telling those businesses simply to “move up the value chain” understates how difficult changing a company’s skills, culture and economics can be. There will be painful restructuring, disappearing roles and some agencies that discover the thing clients were really paying them for was far closer to production capacity than anybody wanted to admit.
But for agencies capable of making the shift, there’s something oddly liberating about the question in the headline. If AI can make the work, then agencies are finally forced to articulate what else they’re for.
The answer can’t simply be more work, produced more quickly. Clients can get that from the technology themselves. It can’t be access to AI, because access is becoming universal. And it probably can’t even be “ideas” in the broadest sense, because generative systems can produce an almost comic abundance of those too.
What remains valuable is the process that turns possibility into conviction: understanding what problem is worth solving, seeing what others have missed, recognising which idea actually belongs to the brand, knowing where to take a risk and where restraint is more intelligent, bringing the right specialists around the problem and being able to explain, when challenged, why this route deserves the client’s money rather than the hundred others that could have been generated just as easily.
That is a more demanding proposition than making the work.
It’s also a better one.
The future agency probably won’t win because it can produce things a client physically cannot produce without it. That advantage is already eroding. It will win because the client believes involving the agency leads to better decisions, stronger work and more valuable outcomes than simply asking the machine and accepting whatever comes back.
And perhaps that’s the answer agencies should become comfortable giving.
If all you need is more content, you may not need us.
If you need to work out what’s actually worth saying, making and spending money on in a world where creating almost anything has become ridiculously easy, that’s where the conversation gets interesting.
