There’s a slightly terrifying question beginning to creep into agency conversations and, unlike most slightly terrifying questions in advertising, this one can’t be solved by adding the word “transformation” to a PowerPoint deck.
If a brand can make the work itself, what exactly is the agency for?
It’s not an entirely hypothetical question anymore. A marketing team can already use AI to develop campaign territories, generate headlines, produce social assets, visualise concepts, resize creative, translate copy, summarise research, create synthetic imagery, storyboard films and produce enough alternative executions to make a creative director develop a nervous twitch. And it can do much of this before an agency has finished arranging the chemistry meeting.
The technology isn’t lurking somewhere over the horizon either. 51% of UK creative businesses are already using AI, compared with 33% across the wider economy, while 99% of advertisers surveyed in recent UK research were engaging with generative AI in some form. This isn’t really a conversation about whether the tools will enter professional creative workflows anymore. They’re already there.
So yes, brands can increasingly make the work.
The mistake is assuming that making the work was ever the only reason they hired agencies.
AI Has Removed One of the Agency’s Oldest Advantages
For decades, one of the simplest arguments for hiring an agency was access. You didn’t have an art director, strategist, photographer, filmmaker, copywriter, motion designer, production company and media planner sitting around waiting for you internally, so you bought access to them externally. The agency was partly an ideas business, certainly, but it was also an infrastructure business. It assembled the people, tools, experience and production capabilities required to turn a business problem into something the public might actually notice.
AI is messing rather dramatically with that arrangement.
A relatively small brand team can now access capabilities that would once have required an entire creative department. What previously meant briefing several specialists can sometimes begin with opening one application. That doesn’t mean the result will be brilliant, of course, but it might be perfectly adequate, and adequate is where the threat starts becoming commercially interesting.
As we’ve explored previously on Creativepool, AI doesn’t have to eliminate creativity to fundamentally alter the creative economy. It only has to start commoditising average creative output. Good-enough copy, serviceable product imagery, competent layouts, respectable social assets and familiar campaign territories are increasingly quick and cheap to generate. When “good enough” becomes a button, an agency proposition built largely around producing more good-enough work suddenly looks considerably less defensible.
Because increasingly, the client can make the stuff too.
Making Something Isn’t the Same as Knowing What Should Be Made
Ramsey Deans
This is where the conversation tends to get muddled. Generative AI collapses the distance between intention and execution. Describe something reasonably well and, seconds later, there it is. That feels enormously powerful because execution has historically been one of the expensive bits.
But when execution becomes almost limitless, scarcity simply moves elsewhere.
A reasonably capable marketing department can now produce dozens, potentially hundreds, of creative territories in the time it once took to brief and review a handful. The problem is that somebody still has to decide whether idea number 37 is genuinely interesting or merely looks like something Nike did four years ago. Somebody still has to understand whether it fits the brand, whether there’s an actual human insight underneath the polish, whether the cultural context makes it smart or catastrophic and whether the most obvious answer is precisely the answer the brand should avoid.
That’s why the most important question in an AI-enabled creative process probably isn’t “Can we make this?” anymore. Increasingly, the answer will be yes. The question is “Should we make this, and why this rather than the other 99 things we could make instead?”
AI dramatically increases the number of possible answers. It doesn’t automatically improve the quality of the question.
Indeed, that distinction may ultimately make genuinely good agencies more important rather than less. We’re already entering an environment where curation can become as valuable as creation, precisely because the ability to generate is becoming ubiquitous while the ability to select intelligently remains scarce. When everyone has more options than they know what to do with, judgement becomes part of the product.
The future agency might therefore spend less time proving that it can execute and much more time proving that it knows which execution is worth pursuing.
Brands Have One Problem Agencies Will Always Exploit: They’re Brands
There’s another fairly obvious reason external agencies continue to exist. Companies spend all day being themselves.
That’s necessary, obviously, but it creates blind spots. Spend five years inside one organisation and certain assumptions stop looking like assumptions at all. They become accepted truths. The customer behaves this way. The product means that. Legal will never allow this. The CEO hates that. We tried something similar in 2022. Our audience doesn’t respond to humour. Eventually, the boundaries become so familiar nobody remembers who drew them.
One of the genuinely useful things an external agency can provide is somebody prepared to walk into that environment and ask why those boundaries exist in the first place. It’s part of what separates a useful supplier from a healthy agency-client relationship. The strongest partnerships aren’t defined by permanent agreement but by enough trust for the agency to disagree when agreement would be easier.
That becomes particularly important in an AI-assisted world because an AI system can be extraordinarily efficient at working within the context it’s given. Feed it the existing positioning, previous campaigns, tone of voice, audience definitions and strategic assumptions and it can become an incredibly productive machine for creating more of the brand that already exists. Sometimes that’s precisely what you want. At other times, it’s precisely the problem.
Daniel Cheng, CEO at Metric Design Studio, makes the case for that external distance particularly well. Having worked on both sides of the client-agency divide, he sees an agency not simply as the party producing the work but as a valuable buffer between an organisation and its own instincts:
“Having worked on both the brand side and the agency side, I can say with confidence that having an agency between the work and the end user is always a good thing, regardless of what created the work. An agency gives the brand room to push past what would normally be allowed internally. It also gives the brand manager a layer of protection when something goes wrong, because there is an agency standing between them and the end results.”
That “room to push” is particularly interesting because some of the best agencies have always functioned as a kind of creative air gap. They allow organisations to entertain ideas that might struggle to survive if they originated entirely inside the corporate hierarchy. An agency can question the brief, challenge category conventions, push against the CEO’s assumptions and present something uncomfortable precisely because it isn’t carrying quite the same organisational baggage.
Cheng’s point about protection shouldn’t be interpreted as agencies existing simply to take the blame when an idea goes wrong. Accountability still matters. It’s more about the psychological and organisational freedom that comes from having an external partner willing to put its reputation behind an idea. A brand manager trying to push a brave piece of work through five internal approval layers is in a very different position when an experienced external creative partner is standing alongside them saying, “Yes, we believe this is worth doing.”
Sometimes a brand needs somebody outside the building to tell it what the building looks like.
AI Gives Brands More Capability. It Doesn’t Necessarily Give Them More Judgement
Steve Rolfe
That distinction between capability and judgement is probably where agencies should be focusing most of their attention now.
Generative technology has made it easier than ever to create words, images, video, music and design. The result, however, hasn’t been an immediate golden age of creativity. It has also been an extraordinary increase in the amount of stuff everybody is capable of producing. More posts, more content, more variants, more videos, more newsletters, more campaign routes and more vaguely cinematic images of attractive people staring meaningfully into the middle distance.
We’ve already looked at why originality becomes more valuable in a world drowning in content, and exactly the same principle applies here. The easier content becomes to produce, the less useful sheer volume becomes as evidence of creative capability. A brand doesn’t gain a meaningful competitive advantage simply because it can now flood the world with 400 additional pieces of content a month. If anything, it becomes more dependent on people capable of working out which pieces deserve to exist at all.
The old bottleneck was production. Increasingly, the bottleneck is attention and judgement.
When producing ten concepts required days of work, the simple act of producing ten credible concepts had value. When an AI system can produce scores of them in minutes, the value migrates towards knowing which ones are derivative, which ones are strategically irrelevant, which ones look seductive but have nothing underneath them and which one contains the small germ of something worth developing properly.
That’s the uncomfortable opportunity for agencies. They have to become much better at being filters rather than factories.
This doesn’t mean agencies stop making things or retreat into an endless fog of consultancy and strategy decks. Craft still matters enormously. In fact, the growing accessibility of competent execution arguably makes genuinely human creativity more valuable, because distinction increasingly comes from taste, originality, cultural instinct and the confidence to move beyond the plausible middle.
The distinction is that craft increasingly sits downstream of a more valuable set of decisions. Which problem are we solving? Which audience tension matters? Which cultural territory belongs to us? Which idea is distinctive enough to survive contact with the real world? Where should AI speed things up and where would it flatten precisely the thing that makes the idea interesting?
That broadly fits the emerging “augmentation first” approach to creative AI, where the technology supports human creative work rather than reducing the entire process to automated production. It sounds less like the disappearance of creative expertise and more like a redistribution of where that expertise is most valuable.
In-Housing Will Accelerate, and Agencies Should Stop Pretending Otherwise
None of this means agencies should get comfortable. Brands absolutely will bring more creative work in-house because of AI, and frankly, they’d be mad not to.
Versioning, localisation, resizing, straightforward social content, internal communications, basic visualisation, presentation graphics and countless other relatively predictable production tasks can increasingly be completed internally, faster and more cheaply. Some brands will build exceptionally sophisticated in-house operations combining permanent creative teams, proprietary brand systems, AI platforms and flexible pools of external specialists. The idea that every piece of communication automatically needs to disappear into an agency for three weeks will become increasingly difficult to defend.
This is one of the central pressures currently facing the traditional agency business model. A client doesn’t actually need to reproduce everything an agency can do internally to change the relationship. It only needs to absorb enough of the repeatable middle that outsourcing every stage no longer makes economic or operational sense.
That doesn’t automatically mean independence from agencies altogether. It means independence from agencies for work the client can now sensibly perform itself.
The threat isn’t necessarily that a marketing department wakes up tomorrow and decides it never needs an agency again. It’s that the department gradually absorbs enough routine capability to alter what it’s prepared to outsource and, just as importantly, what it’s prepared to pay an agency to do.
That’s why determining whether an agency is actually future-ready increasingly comes down to whether it offers something beyond access to production capability. The strongest agencies will still need craft and ideas, but they’ll also need to frame problems, connect creative decisions with commercial ones, manage technology intelligently and adapt their role as their clients’ internal capabilities grow.
Interestingly, the evidence from advertisers themselves suggests AI is already changing agency scopes, team structures and ways of working without simply making agencies irrelevant. Agency value is increasingly shifting towards strategy, effectiveness, integration, governance and commercial expertise as routine execution becomes easier to automate.
In other words, brands may need agencies less often to make the 73rd adaptation of something. They may need them more acutely when they have absolutely no idea what the original thing should be.
The Retainer Full of Stuff Is in Trouble
Chris Hannah
This is where things become commercially uncomfortable.
Agencies have traditionally been able to charge for both thinking and making, and the making was often rather easier to monetise. Hours spent producing campaign adaptations, cutting videos, resizing assets, rewriting copy, building presentations or generating endless variations could all be scoped, staffed and billed. AI compresses those hours, which creates a fairly fundamental problem for an industry still heavily attached to selling time.
If something that once required six people for three days can now be done by two people in an afternoon, the client is eventually going to notice. Trying to preserve exactly the same fee by hiding that efficiency isn’t much of a strategy. It merely postpones the difficult conversation about what the agency is actually being paid for.
That conversation is already pushing some agencies to consider whether the future lies less in the traditional service model and more in building intellectual property, products and systems. Agencies have spent decades talking about the enormous value of their thinking while often pricing that thinking through one of the bluntest measures imaginable: how many hours somebody spent doing it. AI simply makes that contradiction harder to ignore.
That doesn’t mean the answer is for clients simply to pay less for everything. It means agencies have to become much clearer about where their value resides. A strategically brilliant idea doesn’t suddenly become worth less because an experienced creative team reached it quickly, just as a bad idea doesn’t become worth more because 12 people spent three weeks arriving there. If AI removes administrative drag and accelerates production, clients should benefit from that efficiency, but agencies also need commercial models capable of recognising the value of expertise, intellectual property, systems, judgement and outcomes rather than merely counting the hours required to produce an asset.
This could actually force a healthy correction. Agencies have complained for years about being treated like suppliers while often selling themselves through supplier-like units: hours, heads, assets and deliverables. If those units are becoming cheaper and easier to reproduce, the industry has little choice but to make its less tangible value considerably more tangible.
That means showing clients the quality of the decisions surrounding the work, not burying them under process theatre. Agencies capable of understanding the commercial problem behind a brief, challenging weak assumptions and demonstrating how creative thinking connects to results are already better positioned to win more valuable clients. AI doesn’t weaken that argument. It makes it much harder to hide if an agency can’t actually do those things.
Clients don’t need another 90-slide deck explaining the metaphysical importance of a typeface. They do need to understand why the thinking they’re buying is meaningfully better than typing the brief into the same tools everybody else has.
The Future Agency Might Also Be the One That Knows When Not to Use AI
There’s another agency opportunity hiding amongst all the enthusiasm for faster production, and it’s considerably less glamorous than generating a photorealistic film in 14 seconds.
Responsible AI use is complicated.
The difficult questions around copyright, confidential information, training data, synthetic talent, ownership, bias, brand safety, data protection, provenance and disclosure don’t vanish because the output looks impressive. They become part of everyday creative production, and many brand teams simply don’t have the time or specialist expertise to stay on top of every new tool, model and legal grey area themselves.
These issues are now significant enough that copyright, data security, commercial usability, reputation and creative integrity are all being treated as practical barriers to responsible creative-industry adoption rather than philosophical footnotes.
That gives agencies another potentially valuable role. Not simply as enthusiastic operators of whichever shiny platform landed this week, but as experienced partners capable of determining where AI genuinely makes the work better, where it merely makes it faster and where using it introduces more risk than value.
This is why AI literacy is becoming a leadership skill, rather than just an instruction in prompt writing. Leaders increasingly need to understand workflows, governance, provenance and the implications of putting generative systems between a brand and its audience. The clever prompt is useful; knowing whether that prompt should be used in the first place is considerably more useful.
A future-ready agency should therefore be capable of telling a client that a particular workflow is perfectly sensible to automate while another demands greater human oversight. It should understand what data can safely enter a system, how generated outputs should be reviewed and where an apparent cost saving could create an expensive reputational or legal headache further down the line.
Knowing how to use AI will matter. Knowing when not to use it may matter just as much.
Agencies Are Going to Have to Become More Opinionated
Mickey Shu-Ting Chan
For years, plenty of agencies have complained about being treated as suppliers while simultaneously behaving exactly like suppliers. The brief comes in, the work goes out, feedback comes back, changes are made and the whole thing repeats until somebody adds “FINAL_FINAL_v7” to a file name and quietly loses the will to live.
AI can do compliant supplier behaviour exceptionally well. It follows instructions without getting bored, generates options without complaining and doesn’t become emotionally wounded when the client prefers route three.
There isn’t much future in competing with that.
The agencies worth paying for will need to offer something much more difficult: an actual point of view. They’ll need to be capable of challenging the premise of a brief rather than simply executing it beautifully, of spotting the business problem sitting behind the marketing request and, occasionally, of telling a client that the thing they’ve asked for is the wrong thing altogether.
That doesn’t mean being difficult for the theatrical pleasure of being difficult. Nobody wants to pay a substantial monthly retainer to be patronised by somebody in expensive trainers. It means earning the right to disagree by understanding the client’s business deeply enough that disagreement has value.
Indeed, many of the new skills agencies need to build have very little to do with becoming faster operators of AI platforms and considerably more to do with diagnosis, commercial curiosity, judgement and understanding the real problem before anybody begins making things. That changes the agency from an order-taker into something much closer to the strategic partner agencies have spent decades insisting they are.
The arrival of AI only makes that evolution more urgent because everybody is beginning to gain access to broadly similar production capabilities. Owning the tool won’t differentiate an agency for very long. The quality of the decisions made with it might.
“Agency vs AI” Is Probably the Wrong Fight
There’s an irresistible tendency whenever a disruptive technology arrives to turn the conversation into a death match. New thing enters. Old profession panics. Somebody writes a LinkedIn post declaring that an entire industry has 18 months left to live. Everyone argues for a week and then quietly goes back to work.
Reality is generally less cinematic.
Technologies rarely replace an entire professional category neatly. They destroy some tasks, shrink others, transform established workflows and create new forms of value around themselves. AI will unquestionably remove some work from agencies, but it will also remove significant amounts of work inside agencies. Those two things aren’t the same.
In fact, much of what’s emerging looks less like human creativity being switched off and more like creative workflows becoming hybrid human-AI systems. The machine can accelerate exploration, iteration and production while the people surrounding it provide context, intention, judgement, direction and accountability. The precise balance will vary by agency, client and project, but the binary notion that either a human makes everything or AI makes everything is already starting to look rather quaint.
A future agency could require fewer people to produce considerably more, particularly as smaller specialists gain access to capabilities that were once available only to large networks with significant production infrastructure. That doesn’t automatically lead to a barren world in which every CMO sits alone generating Super Bowl campaigns between Teams calls. It could just as plausibly produce a much more fluid ecosystem, where in-house teams handle everyday execution while agencies, studios, production specialists and independent talent are brought in around particular problems, opportunities and moments of genuine creative importance.
That would change what an agency looks like. It may also change the assumption that a single agency needs to make everything for a client simply to justify the economics of the relationship.
And perhaps that wouldn’t be such a terrible thing.
An agency asked to solve fewer but more meaningful problems could potentially spend more time doing the work that justified hiring it in the first place. Less churning, resizing and production bureaucracy masquerading as creative partnership, and more thinking about what the brand actually needs to do next and how creativity might help it get there.
So, Will Brands Still Need Agencies?
Caroline Howden
Yes, but probably not because they need somebody to make things for them.
That part of the equation is changing too quickly. Brands will continue developing their own creative capabilities and AI will make an increasingly large proportion of everyday execution practical to handle internally. Any agency business model dependent on clients remaining technologically helpless is consequently heading for a fairly unpleasant awakening.
What brands are less likely to solve with another piece of software is the problem of perspective. Organisations will still become trapped inside their own assumptions. Marketing teams will still struggle to persuade internal stakeholders to take creative risks. Businesses will still misdiagnose problems, follow category conventions, overvalue what they want to say and undervalue what audiences actually care about. And a machine capable of producing an infinite number of beautifully executed responses won’t necessarily tell you that you’ve asked entirely the wrong question.
That’s where the agency relationship still has teeth. The value sits in understanding the business well enough to challenge it, bringing enough distance to see what the client can’t, applying enough taste to choose between endless possibilities and having enough experience to recognise the difference between something that merely looks like advertising and something capable of actually working.
Perhaps that’s also why the skills required to build a futureproof agency increasingly sit above simple tool competence. The agency that survives isn’t necessarily the one with the biggest AI stack. It’s the one capable of understanding how the technology changes production economics without confusing production efficiency with creative value.
The agency of the future therefore shouldn’t be terrified that a client can suddenly generate 100 ideas. It should be far more concerned if it can’t explain why 99 of them should be thrown away.
AI can already make the work, and over the next few years it will become faster, cheaper and considerably better at doing so. That’s not really the argument anymore. The interesting question is what happens when making becomes the easy bit.
If agencies continue defining themselves primarily by production capacity, the answer could be fairly brutal. But if they recognise that their most valuable contribution has always been the combination of outside perspective, strategic judgement, cultural intelligence, creative nerve and the willingness to push a client somewhere it wouldn’t comfortably go alone, there’s still a compelling reason for them to exist.
Perhaps an even stronger one.
Because in a world where practically anybody can make practically anything, having more output isn’t necessarily much of an advantage. What becomes valuable is the ability to recognise what’s distinctive, what’s relevant, what’s strategically useful and, occasionally, what’s worth leaving completely alone.
AI can make the work. Knowing what work is actually worth making remains a considerably harder problem.
