AI Takes Over Creative Projects. Junior Creatives Face an Uncertain Future.

Two of the world's largest advertising groups are betting their margins on artificial intelligence that doesn't just assist creatives—it manages them. Publicis and Serviceplan are now deploying systems that coordinate entire projects, deciding which AI models to use, maintaining context across iterations, and advancing multiple creative directions without human intervention. The shift from tool to orchestrator marks a fundamental change in how creative work gets done.

The technology, called Luma Agents and launched this week by Palo Alto-based Luma, represents a significant bet that agencies can solve their margin problems by handing off coordination and execution work to artificial intelligence. The timing is hardly coincidental. Publicis announced AI-driven efficiency targets in 2024, making clear its intention to use automation to reduce costs.

Both groups now say they're embedding Luma's technology across strategy, creative development, and production workflows. According to Alexander Schill, Global Chief Creative Officer at Serviceplan, the system allows teams to "develop great work faster" whilst maintaining brand consistency across markets. What's notable here isn't that agencies are using AI—they've been doing that with ChatGPT and Midjourney for two years.

Enjoying this article?

Get stories like this in your inbox every week.

Join the Community

The shift is from humans wielding multiple AI tools to AI systems orchestrating the tools themselves, deciding which models to deploy, maintaining context across iterations, and advancing multiple creative directions simultaneously.

The industrialisation thesis

The strategic logic is straightforward: advertising agencies are service businesses with notoriously thin margins, squeezed between clients demanding lower fees and talent commanding higher salaries. Creative production—particularly the execution work that turns a concept into deliverable assets—is labour-intensive and doesn't scale linearly. If an AI system can handle iteration, file management, version control, and cross-format adaptation, the cost structure improves dramatically.

Luma's pitch centres on what it calls "Unified Intelligence," a model architecture that handles text, image, video, and audio generation within a single system rather than chaining together separate specialist models. The company claims this approach maintains context better than the modular systems built by OpenAI, Anthropic, and others. Its Uni-1 model, described as a "decoder-only autoregressive transformer," processes language and image tokens in the same sequence, allowing it to reason and render simultaneously.

Whether unified architectures prove superior to specialised models remains an open technical question—the AI industry hasn't reached consensus. But from an agency perspective, the appeal is obvious: fewer integration headaches, less context lost between steps, and a single system to train staff on rather than a proliferating toolkit.

Who loses their seat at the table

The more uncomfortable question is what happens to the junior and mid-level creatives who currently do the work these agents promise to automate. Desktop publishing eliminated typesetting jobs in the 1980s. Digital design tools reduced the need for production artists in the 1990s. Those were primarily production-focused roles.

Luma's system claims to handle "conceptual iteration"—evaluating outputs, refining them, advancing multiple directions in parallel. That's the work that has traditionally been how junior creatives learn the craft: executing briefs, exploring variations, developing an eye for what works. If agencies are genuinely moving humans to "taste, direction, and strategy," where does the next generation develop those capacities?

Jain's framing—that "creative teams shouldn't have to spend their time orchestrating tools"—assumes orchestration is drudgery rather than skill development.

Perhaps it is for senior creative directors. For someone three years into their career, it's how you learn to think through a project.

The fine print on intellectual property

Luma emphasises that customers retain "full IP ownership" of outputs, with automated content review to reduce copyright risk and documentation of human involvement. These safeguards matter in an industry increasingly nervous about liability. But the mechanics deserve scrutiny.

The system coordinates across multiple third-party models, including OpenAI's Sora, Google's Veo, and others, each trained on vast datasets of uncertain provenance. When an agent routes part of a project to a model trained on unlicensed imagery, then refines that output using another model's capabilities, who owns what? The legal framework hasn't caught up to these hybrid workflows.

Agencies may find themselves with documentation showing human review but unclear answers about whether their outputs incorporate protected material several steps removed. The requirement for "human review workflows prior to public release" is telling—even Luma's marketing acknowledges these systems can't be trusted to ship work unsupervised. That human checkpoint is the current limiting factor on full automation.

What the deployment really means

When companies say technology is "already embedded" in operations, that typically means pilot programmes rather than wholesale replacement of existing workflows. Agencies of Publicis and Serviceplan's scale don't flip switches—they test, iterate, and scale gradually. The announcement signals strategic intent more than operational reality.

But intent matters. If the largest agency groups are publicly committing to AI-driven workflow automation, their competitors will follow or risk being undercut on price. That competitive dynamic, more than the technology itself, will determine how quickly these systems proliferate.

The creative industries have absorbed previous waves of automation by shifting human labour to higher-value activities. The optimistic case is that this follows the same pattern—creatives freed from execution drudgery to focus on strategic and conceptual work. The pessimistic case is that "taste, direction, and strategy" don't require teams of 50 when the execution scales infinitely.

Agencies will discover which is true over the next 24 months, as pilot programmes either expand or quietly disappear from earnings calls.

More articles by David Adams