If your organisation was already struggling, AI will just make it louder

AI is the ultimate fix: bring in the tools, improve the numbers, move faster. Dr Laura Weis, Global Human AI Strategy Lead at WPP, thinks that framing is the first mistake.

“It makes good working systems better, multiplies value there. But if you have an organisation that structurally isn’t clear, where there’s ambiguity, where the culture isn’t what it should be, there’s no psychological safety, or people are super stretched and inject AI on into these systems, you just increase noise, and they get worse and worse. It’s an amplifier.”

“You see a lot of what I call shadow work: rework that piles up because the first passes aren’t strong enough. You look under the hood and then have to rework a lot. Usually your strong performers get tied into a lot of correction and editing, and that’s not where they’re best placed.” AI polishes the surface, leadership sees the speed, and the strain lands on whoever fixes what’s underneath.

Speed Is Not a Strategy

That’s the problem when AI lands on a broken system. AI gives organisations two things, Laura argues. Speed, which has absorbed most of the attention over the last two years. And space: time to think, to challenge, to connect ideas that weren’t connected before. As Michael Porter said, speed is not a strategy.

At WPP, the internal narrative has shifted over the last year from quicker production to better decision-making. Laura sees the wider pattern: in times of ambiguity, organisations focus on what’s going away rather than what’s emerging, protecting what they have rather than building toward what’s possible. The efficiency question doesn’t go away, but it’s no longer the primary one.

Very few organisations can currently show they’re saving meaningful time while holding quality steady, partly because they’ve put AI on top of old ways of working without redesigning them. “On an individual level, a lot of people are feeling that efficiency, but we can’t scale it.” It’s the gap aibl hears about most often: individuals feel faster, but leaders can’t find the saving in the numbers. The bottleneck isn’t the tool. It’s the decision-making structures and workflows around it that haven’t kept pace.

People are more stressed and burned out than before, not less. Employees feel they need to raise the ceiling themselves, doing more rather than doing better. “It’s really the cognitive load of continuously reviewing AI output and having all that optionality. All of a sudden, to one brief question, if you get a brief from a client and you put it into an AI tool, you have 20 different answers. Like how on earth are you going to make a decision there?” 

At team level, this becomes the paradox of choice: more options, generated confidently by AI, teams increasingly unable to decide between them. Even when a decision is made, the doubt doesn’t disappear. When every alternative looks viable, it’s harder to believe you chose the right one. It’s not just decision paralysis, Laura argues. It’s decision doubt.

Fluency Isn’t Craft

Laura sees organisations moving toward something flatter: smaller pods, outcome-based models, teams working alongside a growing number of agents. The old rules about how people create value won’t survive that shift.

For juniors, the transition looks deceptively straightforward. They’re often already fluent with standard AI tools, no training required, and they bring something leadership needs. “They teach us to not become crusty at the top.” 

But there’s a risk sitting underneath that fluency. “People get to the answer super fast, but they don’t really understand how those answers were formed, when to trust them, and how to really improve them, because they have never done the craft. They’ve never been in the arena, doing the work, learning from what works, what doesn’t work, what fits together, what does good look like, and really experiencing the consequences of when you don’t do good work.”

Her answer, which she admits is rough rather than a finished framework, is an AI driving licence. Generating output and interrogating it are distinct skills. “You need to pair generating stuff with being able to interrogate stuff properly. If you can’t demonstrate that you know how to actually look at AI output and use AI for interrogation as well, then maybe you shouldn’t be using AI yet.” In practice, that means checkpoints built into the tools themselves, “little stop points where you ask: have you actually done the cognitive work to proceed?” rather than a blanket standard applied from outside. It matters especially where tools skip entire chains of thinking that previously had to be done manually.

The longer-term risk: if entry-level work gets absorbed by AI and companies stop developing juniors properly, the pipeline breaks, and rebuilding it later is expensive. Diversity is the less obvious piece. A workforce that doesn’t reflect the range of its customers is less likely to produce work that stands out.

For more experienced people, the gap is different. Deep specialisation still matters, because without craft you don’t know what’s worth connecting. A media planner, for example, might need to build out into strategy, not to become a strategist, but to bridge the two and work effectively across less siloed teams.

What Laura’s research keeps returning to as the defining characteristic is the ability to integrate: “They need to be absolutely amazing at integrating. Connecting the unconnected, across process, across actors, between people and AI, between AI and AI. And when you’re innovating, what you’re really doing is connecting concepts and ideas that have never been connected before.” WPP builds this deliberately, putting people into cross-functional pods where the integration happens through osmosis. 

Automate for Relief, Not Applause

The first question Laura would put to a mid-market business isn’t which AI platform to adopt. It’s what success with AI actually looks like for that business: “figuring out what are the conditions I need to be putting in place for AI to be successful in the first place.”

That means mapping where work currently breaks. “If you just put AI on top of that, it’s not going to fix it. It’s just going to make it worse.” The instinct is also to automate the impressive things first. “Automate for relief, not applause.” Start with the grunt work that drains people and slows decisions, often at the briefing stage or sign-off, then redesign a few of those workflows end to end, with AI embedded rather than bolted on.

From there, define the quality layer: the success signals that matter, and which decisions are crucial. “Don’t do anything with AI if you haven’t really thought about that quality layer.” Speed matters, but the measure is whether AI is helping the business make better decisions where it counts. Clients are asking the same question.

“If you make it about whether AI was used or not, you kind of lose either way.” What clients want to know is whether the work is sound and whether a human was in control. Not passive, not just an approval gate, but actively orchestrating. Accountability has to sit with a person because, as Laura puts it, there isn’t a universally figured out way of holding AI accountable for things. The human layer isn’t a quality check, it’s where responsibility sits.

“Scale what works well, not what demos well. Scale from proof of value, not proof of concept.” And avoid what Laura calls AI theatre: deployments that look good and change nothing that matters.