Infinite creative, finite judgement: what Future of Media London made clear

When we can make anything, deciding what to make becomes the challenge

In his session, Will Akhurst of Brainlabs spoke about us nearing the point where “infinite creative” was possible, where generative AI allows the production of high-quality assets to become faster, cheaper and increasingly unconstrained. As a result, the traditional bottleneck begins to disappear, but what we’re left with is a hard question: what should we actually be creating?

Akhurst visualised the problem as a fitness landscape, a modelling technique typically used in evolutionary science. In this model, each possible execution sits on a landscape, but the most successful executions are represented by peaks in the landscape. AI allows brands to create across far more of the landscape and allows them to iterate at speed to climb the peaks.

The proposed response to this combines two behaviours:

• Exploit known high-performing areas through controlled variations.

• Explore new creative territory through larger, evidence-informed leaps.

What I found particularly interesting was the risk of becoming too good at optimisation. If we continually iterate around what already performs, we may get better and better at climbing one peak without ever discovering whether there is a much bigger one elsewhere. For me, that makes creative exploration just as important as creative optimisation.

What does this mean for brands?

It’s clear the future is not about asking AI to create hundreds of ads. It’s about having multiple, human-built starting points, each of which is deliberately different. From those starting points we can then allow AI to expand, iterate, and test all of those territories rapidly, with human decision-makers reviewing and deciding when to allow that optimisation to continue, when to pause, and when to change direction entirely.

Without strong strategic inputs, AI is just as capable of scaling category conventions, familiar formats and creative sameness. Unlimited production is only valuable if it is creating meaningful variation.

Distinctiveness must come before scale

Wednesday’s Challenger stage opening session from eatbigfish approached the same problem from a different direction: distinctiveness.

Their argument was that challenger thinking is not defined by company size. It is a mindset created when ambition outstrips resources and forces a business to compete differently. The emphasis was not on doing the category basics fractionally better, but on creating a compelling difference and introducing new criteria of choice.

This is particularly relevant in an AI-enabled creative environment. If the starting strategy is generic, faster production gives us more generic work.

This was probably one of the strongest connections I made across the day. We talk a lot about AI increasing creative output, but volume is not particularly useful if everything starts from the same category conventions. The opportunity is not simply to make more creative, but to create more genuinely different starting points worth exploring.

eatbigfish’s wider Cost of Dull work makes a similar point: dull execution often begins much earlier, in safe positioning, category clichés and risk-averse strategic decisions. Differentiation needs to be a strategic decision rather than something we expect creative execution to compensate for later.

This changes the role of a modern creative team. Our value cannot be measured only by how efficiently we turn briefs into assets. We need to help shape the thinking before the asset is made:

• What does the brand want to change?
• What is getting in its way?
• What familiar category behaviour should it refuse to copy?
• What should make someone feel, remember or choose something different?

Human judgement becomes more valuable, not less

The Human Advantage in an Agentic Age, led by Leila Seith Hassan of Digitas and Lin-Sze Teh of Starcom, explored what happens as agents take on more research, analysis and synthesis.

Hassan and Teh spoke about the importance of “Context IQ”: the combination of intelligence and emotional understanding that helps a person recognise intent, cultural meaning and those moments when, despite the available data, “the vibe’s just off”.

I think this is particularly important for creative teams. Creative judgement has always involved things that are difficult to reduce to a score: knowing when something feels culturally relevant, when an idea is trying too hard, or when technically good work simply isn’t very interesting.

AI can process documented knowledge. What AI cannot do is participate in culture in the way a person can. AI does not have lived experience; it doesn’t have empathy and it doesn’t have an instinctive understanding of why people can sometimes make apparently irrational decisions driven by emotional factors.

That does not mean human intuition should sit above evidence. The best work will come from combining different forms of intelligence:

• Data can identify patterns.
• AI can accelerate analysis and iteration.
• Human judgement can interpret context, intent and consequence.
• Diversity of lived experiences must challenge what the system overlooks or flattens.

For me, what was clear here is that cultural curiosity is going to become more important, not less. If everyone has access to increasingly capable tools, knowing what to ask them to make, and being able to recognise when the output actually means something to people, becomes a much more valuable skill.

Agents need more than a task. They need knowledge

An important distinction Jon Block of Syllepsis made in his session was that an AI agent may understand the internet, but it does not automatically understand a particular agency, client or brand.

We saw Block give a live demonstration, in which he gave an agent the same planning task using two different knowledge architectures. The plan changed because the knowledge changed, not because the technology did.

This was one of the ideas from the day that really stayed with me because it made the conversation around AI feel much less theoretical. Agencies already hold huge amounts of valuable knowledge, but a lot of it lives in people’s heads, old presentations, client conversations and the context behind previous decisions.

The challenge isn’t necessarily that AI doesn’t know enough. It’s that we haven’t always made what we know accessible to it.

As Block summarised it: “software depreciates, while knowledge compounds”.

If organisations want agents to produce work that reflects their expertise, it is critical that their expertise is documented. Whether it’s brand principles, audience learning, past campaign evidence, creative decisions or client-specific context, it must all be structured so that both people and systems can use it.

Knowledge architecture is not simply an IT project. It has a direct bearing on the quality of the work those systems can produce.

Delegation requires data and accountability

As the conversation moved from what agents can do to what we should allow them to decide, the topic of data quality and accountability became much harder to separate from the technology itself.

The What Would You Let Agents Decide? panel considered agentic planning, buying, optimisation and reporting, but the underlying issue was the quality and relevance of the data supporting those decisions. The discussion emphasised the need for representative data, statistical confidence, contextual relevance and clear thresholds for autonomous action.

Another crucial question followed: when an agent gets a decision wrong, who owns it?

Personally, I don’t think the interesting question is whether we trust AI or don’t trust AI. That feels too binary. What we need to be asking is: what conditions need to be met for us to be comfortable allowing agentic AI to make decisions without human intervention?

Agents can operate within defined boundaries, but humans still need to decide:

• Which outcome matters?
• What evidence is sufficient?
• What trade-offs are acceptable?
• When should the system escalate?
• Who is ultimately accountable for the result?

What we automate will vary by organisation and task, but the principle will remain consistent: we automate where the evidence justifies it and there are sufficient guardrails in place, not where human responsibility becomes inconvenient.

Creative and media cannot operate in isolation

One of the broader implications of all of this is that the traditional boundaries between creative, media, data and strategy become increasingly difficult to maintain.

Sam Drake’s session made the case that the strongest media plan is not really about the plan itself. It is about who owns the whole answer and whether the client receives one connected response rather than a collection of specialist outputs.

From a creative perspective, I think this becomes increasingly difficult to ignore. If creative is being generated, tested and optimised more dynamically, it makes less and less sense for creative development and media planning to happen as separate stages of a process.

Creative generates signals for media; media performance provides evidence for the next creative decision; audience and cultural insight influence both. It’s vital that knowledge flows through the system and is not lost between teams.

The future will not favour a linear process whereby strategy is handed to creative, creative is handed to media, and reporting arrives after the campaign has finished. The future requires a connected loop, with clear human ownership throughout.

The real risk may be standing still

Now more than ever we are faced with pressure to deliver more with less, and the safest response can often appear to be greater efficiency: producing faster, optimising harder and automating more. However, when everyone has access to the same tools, efficiency alone will not create an advantage.

To think like a challenger, we must have a willingness to depart from convention. If we want to explore creatively, we must accept that not every route we explore will immediately outperform the last. Agentic working will require experimentation, but we need to trust that experimentation process, within clearly defined boundaries.

As David Meikle discussed, it’s vital that we reframe risk, no longer seeing it as recklessness, but instead seeing it as an opportunity for change, achieved through conscious choices, guided by appropriate guardrails, while allowing us to learn quickly enough to act on the result before the landscape changes.

In summary, the lesson I took from Future of Media London is that AI is not going to remove the need for creative leadership; it’s going to shine a spotlight on the quality of creative leadership. We are at the cusp of a world where production is abundant, and judgement is scarce. Execution is being automated, and as a result accountability matters more than ever. Every brand has access to similar tools, so the advantage has to come from the knowledge, perspective and courage behind them.