AI in Advertising Creative Work: What Agencies Are Replacing, What They Are Not.

The Production Layer Is Going First

Stock photography licensing is the clearest casualty of AI's entry into advertising production. A single rights-managed image cleared for multi-market use can cost between £300 and £800, sometimes more, depending on territories, duration, and media channels. AI image generation via credit-based platforms — Midjourney, Ideogram, Flux 1.1 Pro — brings that per-image cost down to roughly US$0.02–0.10 once you amortise subscription or API credits. More importantly, it covers brief-specific imagery that generic stock libraries never could: a product shot in exactly the right context, with exactly the right demographic, in the exact visual register the campaign requires.

Ideogram 1.0 and 1.5 position themselves explicitly for logos, posters, packaging, and ads, with typography control that was a persistent weakness in earlier image generators. Flux 1.1 Pro, available via credit-based platforms and APIs, pushes towards 4K-class stills with cinematic lighting suited to product and lifestyle work. These are not hobbyist tools. Agencies are running them in production.

The same logic applies to copy. Generating fifty A/B variants of a paid-social headline, or producing localised CTAs for a pan-European performance campaign, is now a sub-dollar task using Claude 3.5 Sonnet or GPT-4.1 via API. The token economics are stark: Claude 3.5 Sonnet runs at approximately US$3 per million input tokens and US$15 per million output tokens. Hundreds of copy variants cost less than a round of coffees. The real cost is not generation — it is the human reviewing, selecting, and editing, which is where the time actually goes.

Motion graphics for lower-funnel ads are following the same curve. Animated product callouts, branded end-cards, and short-form explainers that would previously have been commissioned from boutique studios are being produced in-house using tools like Runway Gen-3 Alpha, which generates five to ten second clips at up to 1536×864 from text or image prompts, with plans in the region of US$15–35 per month for several thousand credits. At a converted cost of roughly US$0.10–0.50 per second of output, the economics versus a commissioned social asset — which might run £1,000–£5,000 for a six to ten second piece — are not comparable. Voiceover for regional campaigns is taking the same route: tools like ElevenLabs and Descript Overdub handle multilingual VO at well under US$1 per finished minute at scale, removing the per-locale talent booking that previously made regional campaigns expensive to produce.

Amazon's Creative Agent, now in beta in Creative Studio, makes the pattern explicit: it handles research, concepts, multi-scene video scripts, image generation, animation, voiceover, and music for Amazon placements, with no additional tool cost on top of media spend. The infrastructure is already there for the brands willing to use it.

The common thread across all of this is not creativity. It is specification. Repetitive, spec-driven output with low conceptual weight — image variations, CTA swaps, end-cards, localised VO — is credit-priced infrastructure now, not brief-by-brief outsourced craft. Agencies are not replacing creative thinking. They are replacing volume production.

What AI Cannot Replace in an Agency Workflow

A language model can generate fifty taglines for a brief in under a minute. What it cannot do is tell you which one is right for this brand, at this cultural moment, given what happened to that brand's reputation six months ago and the client relationship your strategy team has spent two years building. That distinction matters enormously in advertising, where the wrong idea executed brilliantly is still a failure.

Brand strategy and campaign positioning remain human-led. BCG's analysis of AI's role in modern advertising is direct on this point: as generative AI becomes a native advertising surface, the complexity of where and how a brand intervenes in that landscape increases, driving demand for strategy and planning talent rather than reducing it. AI is being used as a thought partner in concepting — generating routes, testing angles, stress-testing positioning — but it is not making final strategic calls. The creative director who kills a strong execution because it serves the wrong strategy is doing something AI cannot replicate: exercising taste in service of a business outcome.

Client relationship work sits entirely outside current AI scope. Reading the room in a pitch, sensing that a client's stated objection is not their real one, knowing when to push back and when to defer — these require accumulated context, trust, and social intelligence that no model can substitute for. The same applies to legal and compliance review. In regulated categories — finance, pharmaceuticals, alcohol — no regulator in major markets currently accepts AI-only review of creative materials. Qualified humans own that process regardless of how the asset was produced.

Conceptual originality under a tight cultural brief is a harder question, and the honest answer is that the boundary is contested. But the work that demands genuine novelty — the idea that reframes how a category talks about itself, or lands a cultural insight that feels genuinely surprising — is still a human responsibility. AI synthesises patterns from existing output. That is useful for exploration and useful for volume. It is less useful when the brief demands something that does not yet exist in the pattern set.

Where Agencies Are Quietly Expanding Spend

The narrative that AI compresses agency headcount misses what is actually happening to budgets. The work is not disappearing. The shape of it is changing, and agencies are spending in new places to keep up.

Prompt Engineering and AI Art Direction

Getting consistent brand outputs across dozens of generated assets is a skilled job. Agencies are beginning to formalise it. Job boards since late 2024 show increasing listings for GenAI Creative Lead, AI Art Director, and Creative Technologist roles — people who understand both the creative brief and the model's behaviour well enough to bridge them reliably. Tools like AdCreative.ai, which markets itself as a creative OS layer that takes in brand assets and outputs channel-specific creatives at volume, reinforce the point: someone has to own the prompts, the guardrails, and the quality assurance. That person is a creative, not a developer.

Creative Technology and Workflow Design

Building the pipelines that connect briefs to outputs to approvals is a growing internal investment. BCG's framework for AI-native media explicitly recommends that marketers "build the operating system for AI-native media," centring workflow design, data hygiene, and rapid experimentation as core capabilities — not tools you subscribe to, but internal infrastructure you design and maintain. Agencies that have built this are outperforming those still assembling it ad hoc per campaign.

Strategy and Planning

Strategy headcount is holding or growing at most holding companies. Clients need more help interpreting a media landscape that AI is actively changing: new surfaces, new attention patterns, new questions about where brand money actually works. That context-reading is a human service, and the demand for it has not shrunk because image generation got cheaper.

Quality Control and Brand Governance

Generative output at volume requires systematic review before it reaches clients. What was once an assumed responsibility — someone would catch the off-brand output before it shipped — is becoming a formalised function with dedicated ownership, documented standards, and structured approval gates. Agencies that skip this are the ones whose AI-generated campaigns surface in trade press for the wrong reasons.

Multi-Model Consolidation

Enterprise buyers are consolidating rather than expanding their tool stacks. The pattern across agencies is consistent: one platform with multiple foundation models behind a single credit system, rather than separate vendor contracts and governance per tool. Managing a Midjourney subscription, a Runway account, a Claude API key, an ElevenLabs seat, and a web builder as five separate relationships is not efficient at agency scale. Platforms that consolidate these into one workflow with one credit system reflect how agencies are actually buying now.

The Economics: What the Maths Actually Looks Like

The savings are real. The framing of them matters.

Asset type	Traditional cost (approx.)	AI production cost (approx.)	Where the saving goes
Single stock image (multi-market licence)	£300–£800+	US$0.02–0.10 per image	Review, iteration, brand QA
6–10s social video asset	£1,000–£5,000	US$0.10–0.50 per second (Runway Gen-3)	Creative direction, approval
Multilingual VO (per locale)	£200–£600 per talent booking	Under US$1 per finished minute at scale	Script editing, tone review
Hundreds of A/B copy variants	Junior writer day rate x multiple rounds	Under US$1 in API tokens	Selection, editing, final approval

The saving is not a pure profit transfer. The time cost shifts from commissioning to prompting, reviewing, and iterating — which requires a skilled operator who understands both the brief and the tool. An agency that replaces a production budget with a platform subscription and assumes the rest handles itself will find the output is faster and worse. The value is in the combination: cheaper generation plus better creative direction, not generation without direction.

Junior roles are being restructured, not simply eliminated. The expectation at an increasing number of agencies is that a mid-level creative now delivers output that previously required a team of three. That changes hiring ratios. It does not collapse headcount overnight, but it does mean that the next junior hire is measured against a higher productivity baseline from day one.

Billing models are under pressure. Agencies that bill on time-and-materials face a direct problem: AI-assisted production compresses hours, which compresses invoices on the same scope. The agencies absorbing AI costs into their internal tooling and billing clients on fixed-fee or value-based contracts are protecting margin by decoupling revenue from execution hours. WPP and Publicis have both addressed this in earnings commentary; BCG's recommendations to adopt new commercial models in AI contexts reflect the same structural shift.

What This Means for Creatives Working in Advertising Now

Speed was always a competitive advantage for junior creatives. AI has taken speed off the table as a differentiator. The skill that protects a creative career in advertising now is quality of judgement: knowing what to make, why, and for whom.

Creatives who can direct AI output — evaluate generated assets critically, iterate them to a professional standard, and know when to discard a technically competent result that is strategically wrong — are more valuable than those who can only produce by hand or only prompt by instinct. The combination matters. A graphic designer who can take a brand bible, build a Midjourney prompt set that produces on-brand outputs consistently, and QA those outputs against campaign objectives is doing something more sophisticated than either pure manual production or undirected generation.

Understanding the full pipeline is now a competitive advantage even for specialists. A motion designer who knows how performance data flows back into creative briefs, or a copywriter who understands how their social variants will be tested and what metrics determine which survives, is contributing at a level above their nominal specialism. Agencies are watching which creatives demonstrate this fluency and which treat AI tools as someone else's problem. The observation is increasingly a factor in performance reviews and hiring decisions.

The professionals most exposed are those whose value proposition was primarily speed or volume. That is precisely what AI addresses first. A junior retoucher whose role was to produce image variations at pace, or a writer whose main output was tactical copy at volume, is working in the part of the production stack that is being automated most directly. The response is not to compete with the tool on its own terms. It is to move up the value chain: into direction, into strategy, into the kind of critical taste that makes generated output useful rather than merely adequate.

That move requires deliberate investment. Creatives who are waiting for AI to stabilise before engaging with it are accumulating a deficit that is compounding. The ones who are building fluency now — not necessarily expertise in every tool, but confident working knowledge of generation models, prompt craft, and iterative review — are building the professional moat that will matter most over the next three to five years.

Where Stensyl Fits Into an Agency Workflow

The consolidation pattern described above — one platform, one credit system, multiple model types — is exactly what Stensyl is built around. For an agency creative or marketing team, it means generating campaign visuals, writing copy variations, producing audio for social video, and building a client-facing microsite without switching between five subscriptions or managing five separate vendor relationships.

The Marketing surface consolidates social posts and performance ad formats — carousels, ad-specific layouts — with research-backed copy generation in one studio. A performance marketer building a paid-social campaign can move from brief to copy variants to formatted visual assets without leaving the environment, which removes the handoff friction that typically slows production-heavy campaigns.

The Write surface with its multi-model picker lets strategists and copywriters run the same brief through different models and compare outputs directly. When a client has a distinctive voice — dry, technical, category-specific — the ability to compare Claude Sonnet 4.6, GPT-5.5, and Gemini Pro on the same input makes tonal matching a practical step rather than a guessing game. The Canvas node-based editor extends this further: a prompt can flow into image generation, then into a graphics layer, then into copy, all inside one templated workflow that can be reused across a campaign family rather than rebuilt each sprint.

For visual work, the Image surface spans over twenty models, which means an art director is not locked into one generation style. The Motion studio handles motion graphics export for the kind of animated lower-funnel assets described earlier. The Audio surface covers voice, music, and sound effects, which means localised VO for a regional campaign and the Web surface handles landing pages and microsites for campaign-specific builds.

Credit-based pricing across all surfaces means an agency can scale usage up during campaign sprints and down in quieter periods without paying for idle seat licences. Plans run from £10 to £84 per month depending on credit volume and concurrency needs, with a free starting tier that includes 150 credits and one video render for teams evaluating the platform before committing. The model picker is the same on every plan: Claude Opus 4.8, Claude Sonnet 4.6, GPT-5.5, GPT-5.4 mini, Gemini Pro, and Gemini Flash are all available regardless of tier.

BCG's framing of this infrastructure category as "building the operating system for AI-native media" captures what platforms like Stensyl are trying to provide: not a single tool that does one thing well, but a connected environment where the brief, the generation, the iteration, and the client-ready output all live in one place, auditable and repeatable at campaign scale.

---

The advertising industry is not in an AI moment. It is in an AI restructuring that will take several years to settle. The production layer is already transformed. Strategy, relationships, and creative judgement are not next — they are the durable centre. The professionals and agencies that understand which is which, and invest accordingly, are the ones who will be standing at the end of it.