Every agency AI announcement this month was about making things faster.

Not one was about changing what the customer actually touches.

Creative Boom's September industry roundup is a decent snapshot of how the creative industry is responding to AI right now. SuperHeroes appointed a creative director to lead AI and CGI. LERMA/ hired the former AI solutions director from Monks as head of production, to bring production thinking earlier into the creative process. M+C Saatchi Talk packaged PR, search and AI into a new framework. Sensible moves, made by good agencies.

Look at the shape of them, though. Every one is production-shaped. AI as a better camera, a better render farm, a faster copy machine. The deliverable at the end is unchanged: a campaign, an asset, a film. It just arrives sooner and costs less to make.

SuperHeroes themselves gave the game away, with admirable honesty: "AI has made it easier than ever to make something. It hasn't made it easier to make something people care about."

Production was never the bottleneck. So why is production where all the investment is going?

When everyone has the same tools, where's the advantage?

Here is the uncomfortable economics of the efficiency play. Gains from tools everyone can buy do not stay in your margin. They get competed away.

When every agency can produce the campaign in half the time, clients simply expect the campaign in half the time, at half the fee. Faster output becomes the client's expectation, and then the client's discount. Worse, the whole pitch is framed in cost terms, which is an open invitation for procurement to reprice you. You have handed them the argument.

Production AI is table stakes. You need it to stay in the game. But it defines the floor of the market, not the ceiling, and a strategy built on it is a race to the bottom that someone with lower overheads will always win.

The question worth asking is not "how do we make things faster with AI?" It is "what should we be making now that AI exists?"

Whilst agencies sped up production, the interaction layer got rebuilt

Here is what happened in a single week in September, while the industry was announcing AI creative directors.

Amazon rolled out agentic AI to its third-party sellers and reported that sellers now accept its AI assistant's recommendations more than 90% of the time. Payments and infrastructure players including Alchemy, Worldline and Noibu shipped integrations that let AI agents discover products, complete purchases and fix ecommerce issues autonomously. Meta's Muse assistant topped the free app charts in the US and Canada, added outbound calling, and started phoning venues to make bookings on customers' behalf. Some venues responded by banning bot reservations. HubSpot made AI agents core workers inside its CRM and watched agent actions grow 3.5x.

Read that list again and notice what it describes. It is not faster production of the old experience. It is the customer interaction itself changing hands. People are not asking for a better booking flow. They are hiring an agent to skip the flow entirely. That is not a stated preference in a focus group. It is revealed preference, at app-chart scale.

The interaction layer of commerce and service, the thing digital designers have spent twenty years crafting screen by screen, is being rebuilt around AI on both sides of the counter. And the creative industry's response, on the evidence of its own trade press, is to make the old deliverables more efficiently.

The questions nobody is asking

The advantage in this era does not lie in producing the current experience faster. It lies in deciding what the customer interaction should become. These are the questions I am not hearing asked:

Should this booking flow become a conversation? Not "should we add a chatbot", which is the 2023 question. The 2026 question is whether the interaction itself, the sequence of steps a customer walks through, should be replaced by a dialogue that holds context, handles exceptions and completes the task.

Should this product page be agent-legible? When a growing share of your customers arrive as an AI agent acting on someone's behalf, your beautifully art-directed page is invisible. What matters is whether your inventory, pricing, policies and availability are structured so an agent can read, negotiate and transact. Being agent-legible is a design problem. Blocking the agents, as the restaurants are discovering, is a holding action, not a strategy.

Which of this brand's service interactions should move to an AI-native model, and what is each one worth? Not everything, and not all at once. Each candidate interaction has a value, a risk and a customer tolerance attached. The engineers building the agent-commerce rails already know this. Their published guidance this month was to score each autonomous action by severity, reversibility and customer exposure. That sentence is a design framework wearing an infrastructure costume, and right now no design team owns it.

The platforms are answering these questions. Generically.

Here is the urgency, and it is not the usual "adapt or die" hand-waving.

The software platforms are not waiting for the design industry to catch up. Salesforce, Amazon, Meta, HubSpot and the rest are answering the interaction questions right now, at scale, in code. The catch is that they are answering them generically. Every brand gets offered the same off-the-shelf shopper agent, the same default conversation, the same template experience. Efficient, capable, and identical.

If that sounds familiar, it should. It is the same dynamic that gave every brand the same website template in 2005 and the same app patterns in 2015, until design showed up and made differentiation the point. The historical job of design has never been to invent the technology. It has been to take a generic capability and make it distinctively, recognisably, commercially someone's.

So the open question is not whether customer interactions will be rebuilt around AI. The platforms have settled that. The question is whether anyone brings brand judgement and customer specificity to the rebuild, or whether every brand ends up with the same default agent. That is not a gap in the market. That is the job.

Migration, not transformation

None of this requires a grand transformation programme, and I would argue it actively punishes one. The way this work happens is one interaction at a time.

Pick a single high-value customer interaction. Score it: what is it worth, what breaks if the AI gets it wrong, how reversible is a mistake, how exposed is the customer. Redesign it as AI-native. Ship it, measure it, and attach a commercial number to the result. Then let that number decide what migrates next.

This approach has a property the efficiency play lacks: it compounds. A faster render pipeline produces the same asset next quarter. But "we migrated the booking flow to conversation, measured what customers actually did, and now know which interaction to move next" builds a proprietary map of a brand's real customer behaviour. Every migration makes the next one better informed and harder for a competitor to replicate. Efficiency is a one-off gain. A migration sequence is an annuity.

Design has the weight to lead this

One more thing from the same September news cycle, easy to miss amid the AI noise. The Design Council reported that design now contributes £136.7bn to the UK economy, up 40% since 2019, making it a bigger contributor than retail.

That figure covers the whole design economy, not digital product design alone. But it settles an old argument. Design in this country is not a downstream, decorative function that gets called in to polish what others have decided. It is one of the economy's larger pillars, and it is growing faster than the sectors it serves. It has the economic weight to lead structural change, not just to style it.

Which makes the current moment a strange one to spend on speeding up the production of yesterday's deliverables.

The opportunity of this era is not designing with AI for the past. It is designing for the AI era: deciding, brand by brand and interaction by interaction, what customer experience becomes when the customer might be a person, an agent, or a person directing an agent. The tools are the same for everyone. The judgement is not.

That is where the advantage lies.