Operating AI at Scale, Preserving the Work - with Zoe Scaman and Tim Rodgers

Open LinkedIn and the mood is predictive: agencies on notice, AI advancing, little certainty. A post circulating this week told advertising workers, without hedging, that jobs will go and hiding won’t help. It struck a chord as the numbers grow harder to dismiss. The post urged people to be “unignorable.” Justin Billingsley, a former Publicis Groupe CMO, echoed it to the holding companies’ roughly 480,000 employees: read it, feel it, act on it.

At this year’s Adobe MAX conference, a global survey of 16,000 creators reported that 86 percent already use generative AI in their workflows (meanwhile, Canva’s free, unified Affinity suite challenges Adobe’s model; we’ll tackle that separately). The World Economic Forum projects labour-market churn of about one-fifth of current jobs by 2030, roughly 170 million roles created and 92 million displaced, and says 39 percent of today’s skills will be reshaped or obsolete. Nearly half of employers plan to reorient around AI; two in five expect reductions where tasks can be automated.

Against that backdrop, I spoke with Zoe Scaman, global keynote speaker and founder of Bodacious, a strategy studio focused on culture, fandom and organisational design; and Tim Rodgers, founder of Ace Workflow™ and an Emmy and Cannes Lions winner whose team implements agentic-AI operations for clients including Google, Disney, Ryan Reynolds’s Maximum Effort, Paramount and AWS.

What follows is a reported conversation on change, differentiation, talent, guardrails, and the work that must remain human.

What leaders are getting wrong

The reflex, Tim Rodgers argues, is to chase showpieces before fixing how the work actually moves. “I think a lot of CMOs are going to image generation and video generation tools. Obviously AI is a lot of things, but a lot of teams are going to these flashy toys… meanwhile their operations are a mess and they've got teams copying and pasting from one system to another. There are other AI tools that can bring a lot of efficiency and allow people to focus on the work.”

The hype cycle accelerates the misread. “It's a very clickbaity situation…A lot of this is posted on LinkedIn and people are trying to get likes and follows. The biggest thing is a lot of people haven't really played with it personally and figured out what it is and what it isn't, so there's a lot of reliance on other people.”

Zoe Scaman places the same impulse in a harsher financial frame. “The conversations I'm having at the moment with c-level people are that they are under an enormous amount of pressure when it comes to shareholder returns…share prices...they are basically saying, ‘how do we cut cost as quickly as we possibly can?’” The mandate follows: “They're basically saying, ‘We want to implement AI as fast as we can to increase efficiencies, to increase productivity, to look at how we can basically cut headcount. How can we increase speed across all of these processes? How can we do more with less?’ That is their focus at the moment.”

Her caution is blunt. “Everyone's trying to get after these gains, which is fine because that is the first baby step when it comes to AI exploration: how can we save money, how can we save time, how can we save headcount. But if you just stop at efficiency and cost cutting you're in a race to the bottom because very quickly you're going to become commoditized. Everyone's got the same tools, the same processes, the same pitch, ‘we can save you money, we can save you time’, but there's no real differentiator after that.” The harder question, she says, is parallel: “Once you've done the foundational stuff, in parallel how are you creating your differentiator? What makes Publicis different from WPP, different from Dentsu, and others? How are you bringing creativity to the fore? What is your version of creativity in this AI era? What are you going to do with talent? Why would talent join you when everyone's freaking out that you're going to cut their jobs or turn them into a robot assistant. Efficiency is great. Tick the box, get that done at the beginning.”

Where differentiation actually comes from

With access to similar frontier models, Rodgers argues the edge sits elsewhere. “The point is we all have access to the same models. How do we get better output from that? We can prompt, but the biggest differentiation is data.”

His first suggested move is simple: capture what your organisation already knows. “One of the easiest turnkey ways of collecting data is transcriptions. We do this a lot with our clients, start to transcribe client calls. Transcribe internal calls. If we can start to do that, we can build up a dataset that's unique to that agency. How does that agency talk about strategy or what is a good idea? All of that information is in those conversations.” The point isn’t storage; it’s use. “We can collect those, then train models to help teams come up with better ideas because we can help push the ideas along. We can use agents to review and improve and iterate on the ideas that are coming out.”

The leak he sees everywhere is wasted ideation. “If you're doing a creative pitch, you might come up with 20 ideas. You might only pitch two of them and the other 18 ideas disappear, which is even worse if you've got freelance talent, it walks out of the building.” His teams start by stopping the bleed. “A lot of the tools we're creating are about first collecting the data and then providing very turnkey ways for individuals to access that information. I think data is going to be the only differentiator between an agency from the perspective of an AI tool, and then obviously your talent on top of that. Approaching it like ‘oh we just purchased Chat Enterprise’ is not going to cut it.”

The talent market: hemorrhage, fluidity, machine scale

Scaman expects a shakeout, and sees early blueprints for what replaces the current headcount map. “We are going to go through a massive hemorrhaging of talent, but I also think it's necessary. The industry has become bloated, very process, paint-by-numbers focused. We are producing mass mediocrity.”

What comes next, she says, depends on the “architectures” agencies choose. “It’s going to be based on the different architectures agencies want to build when it comes to talent… David Droga just stepped down as CEO of Accenture Song… he said all of the gray mediocre talent in the middle is going to be gone… AI is going to become a superpower for incredibly talented people.” Elsewhere, she notes, fluid networks are the bet: “Publicis built Marcel… basically their AI talent engine… increase fluidity of talent across the portfolio… skills and availability in real time… come together around projects; when the project was delivered they would disband.” And another path is sheer scale and process: “Then you look at WPP… making big deals with major AI companies…‘we're going to make everything super efficient, automate everything, everything process-based.’ Indirectly telling talent: you can come here and train our machines, work with the robots.”

The choice for individuals is immediate. “If you are top talent, you need to decide now: where do I invest my time and energy to grow my career and portfolio? … Ultimately there's going to be a massive cut of talent, and the talent left has to decide where they want to go and how they want their career to progress based on these positionings.”

On the brand side, Rodgers sees in-housing accelerating. “There's a huge movement of talent from agencies to in-house teams. We're seeing the rise of the in-house agency. I've been to three conferences this year solely focused on in-housing, and there's a big focus on awards so they can attract the best talent.” The advantage, he says, is fidelity and speed. “The beauty about an in-house team, going back to the data point, an in-house team can train their systems to a much higher level of fidelity because it's a brand or a suite of products… You can really fine-tune from the efficiency perspective… and from a creativity perspective you can train models on your customers, your personas… train the model on the brand, the tone of voice.”

He argues the content treadmill demands a different engine. “Over the years we've built up human teams to fight an algorithm, which doesn't make sense. We're seeing in-house agencies move to a model where there is an AI-orchestrated content creation process automatically finding insights, writing briefs, creating content. It's all human-signed-off and orchestrated, but the majority of the work is created by agents.” And the groundwork is often missing. “Across the board… clients are using a Google Sheet tracker to try to manage a hundred million dollars of creative marketing projects every year… There's no visibility across projects. We're mostly working with teams inside enterprise organisations to get the foundational basic things right. From there you can build a model or an AI agent very easily because you've got your data, single sources of truth, it's accessible… Brands and clients are nowhere near as advanced as they say they are.”

Guardrails: don’t let the muscles atrophy

Scaman’s sharpest concern is cognitive drift when teams over-delegate to machines. “One of the most important ones for me is making sure you've got guardrails in place around augmentation versus atrophy, which I've written about extensively. What we're seeing more and more is people who lean into AI tend to get it to do the entire thinking process for them. They start to lose the ability to think in a certain way, to write a creative brief, to do very basic thought processes in their own mind.”

The risk, she says, is highest for juniors. “I'm really worried about this, especially with junior talent… They're coming straight into an agency in which AI is encouraged… We shove them straight into an AI-generated workflow and say good luck.” Her practical hedge: design intentional gaps. “You might want some AI-free concept sessions to stretch those muscles… How do you make sure people don't become too individualized in their processes? One of the most powerful things of agencies is having a mix of people in a room bouncing off one another… The way we set up AI tools at the moment is very solitary: your own screen, your own workflow… that atrophies social cohesion.”

None of this is anti-tool. “It can be an amazing tool when it's wielded with intention… we still want social cohesion… we still want you to ideate without these things and not become over-reliant.”

IP and the “brain in the machine”

On ownership, Rodgers is pragmatic. “Most contracts with employees give the employer all IP rights… we work with a lot of creatives and individuals to help them understand they can create their own datasets as well… by training a model on that, you've got an assistant for your way of writing.” And most value, he says, will be infrastructural. “A lot of AI, as we've used it for the past 10 years, will be hidden, behind the scenes, helping the user get access to the right information to do the task… For media companies we build a tool that ingests an RFP from a client and automatically matches it to available media opportunities… I don't think the value is in an individual's prompts… have [a call transcription] summarised with key points and delivered via email or Slack… That's a better way of working.”

Scaman worries about what workers leave behind. “When you're getting people into your agency and you've got a centralized agency tool… the expectation is for every person to use that tool and therefore upload their work processes, their IP, their brain into it… If someone leaves or gets fired, essentially their brain has been uploaded into the system and contributes to the compound knowledge… They've essentially got a virtual copy of me in perpetuity.” The law is unsettled. “There is so much happening in the courts around copyright… Australia… Anthropic… ‘Fine, we'll pay you three grand a pop’… the judge said, ‘No, that's not enough.’… I don't know how we track our contribution… whether there's an option to have a right to be forgotten… It's messy… People need to be wary of what they upload and use.”

Her line on ideation is balanced but firm. “From a friction point of view, I do agree some ideation should remain human. But I don't agree all ideation should remain human because if you can wield these systems with intention, and in a way that turns you into someone who can think more expansively and in depth, they're incredibly powerful levers for ideation.” The non-negotiable piece, for her, is curiosity. “It goes back to Russell Davies' book ‘Do Interesting’… notice, collect ideas, connect disparate pieces… The uniquely human piece is your voracious appetite… A large language model is… predicting the next right answer… built against originality, novelty, newness… We can't outsource our ability to connect different dots to a system fundamentally built to do the opposite.”

Conclusion

In my first year of university, Etherpad, later acquired by Google and folded into the live co-editing that now defines Google Docs, changed how I worked: faster drafts, cleaner edits, real collaboration. Not AI, but the same thread: tools set the tempo; people decide the meaning.

That runs through this debate. Zoe Scaman argues for protecting judgment, taste and the social exchange that strengthens ideas, and warns of skill atrophy when teams over-delegate to generative AI. Tim Rodgers argues for engineering the spine, AI at scale: data, workflows, governance, so work can move without breaking. If efficiency outruns distinctiveness, the work flattens. If distinctiveness lacks an operational frame, it stalls.

A workable programme looks like this: use AI to strip repetitive tasks; reserve human time for decisions that require taste and accountability. Build first-party data and shared ontologies so models learn your context, not a median. Set guardrails, consent, IP, evaluation sets, cost controls, so scale doesn’t erode trust or craft. Create incentives for cross-team craft and measurable outcomes, not volume. Keep deliberate friction where it sharpens thinking (briefs, critiques, live co-creation); automate the rest.

Three questions to pressure-test your posture:

1. What unique data do we own, and are we actively learning from it?
   
   
2. Where must humans stay “on the hook” for quality, and is that time protected?
   
   
3. How will we track efficiency gains and creative differentiation on the same dashboard?

The summer before I began university, I reread John Steinbeck’s Of Mice and Men. It isn’t a metaphor for this moment, but the times brought it back. The small lesson I kept: when economics shift, look hard at who is left out, and build the structures that hold them.

Build the rails so the work can move; keep the standard that makes it worth moving.

‍