Yes, you’re more productive. Yes, your consumers might be spending less. Both can be true.
The economics of AI are splitting into competing narratives. Both have consequences for how brands plan, spend and build teams.
In late February, Citrini Research published “The 2028 Global Intelligence Crisis” - a scenario piece modelling what happens if AI-driven productivity growth runs ahead of the labour market’s ability to absorb it. It went viral in finance circles, prompted lengthy debate on Ark Invest’s podcast and fed into a wider conversation from Dario Amodei, Andrew Yang and others. Markets moved on. But the questions aren’t going away - and they have direct consequences for how brands plan, spend and build teams.
The tension we need to work through comes as two very smart groups of people look at the same data and arrive at completely opposite conclusions. On one side: AI drives productivity growth, GDP expands, capital recycles and the labour market adapts as it always has. On the other: productivity soars but gains pool at the top, consumer spending hollows out and we end up with what Citrini calls “Ghost GDP” - output that shows up in national accounts but never circulates through the real economy. Machines spend nothing on discretionary goods.
It seems both sides agree on the diagnosis. The disagreement is about the prognosis. For the business planning cycle, the gap between these scenarios is a gulf between investing for growth on one hand, or preparing for structural demand erosion on the other.
What follows is an attempt to lay out both positions plainly and land each on what it means for brands and communications. Because nobody knows. But you still have to plan.
1/ Will GDP growth offset job losses?
The most credible voice on the pessimistic side is the person building the technology. Dario Amodei, CEO of Anthropic:
“My view is the signature of this technology is it’s going to take us to a world where we have very high GDP growth and potentially also very high unemployment and inequality. We’ve never had a technology that’s this disruptive. So, the idea that we could have five or 10% GDP growth, but also 10% unemployment, it’s not logically inconsistent at all.”
The optimist case comes from Ark Chief Futurist, Brett Winton. Surplus capital doesn’t vanish - it cycles into infrastructure, data centres, new ventures. The remaining human labour becomes scarcer and therefore more valuable because technological deflation brings costs down. GDP growth + deflation = rising living standards, even with fewer traditional jobs.
It’s a clean argument. It assumes the money moves fast enough and distributes broadly enough to prevent a consumption crisis in the meantime.
Alap Shah, co-author of the Citrini scenario, asks the question underneath:
“What’s the marginal propensity to consume for someone with $250 million in assets who gains another $100 million, compared to a paralegal who just lost their job?”
Andrew Yang, who has been talking to CEOs quietly replacing departments says:
“You have a department of six people, let’s say coders or designers, and then you end up eliminating four or five of them. And then there’s one designer-coder empowered by AI.”
Good for margins. Less obviously good for the consumer economy those margins depend on.
What this means for planning: If Ghost GDP is even partially real, consumer spending is likely to continue diverging (yes, that’s a polite term for ‘even more inequality’). Premium categories hold up. Mass-market faces structural pressure. Sadly, we’ve probably seen this movie before and should expect a consumer base that’s smaller and more polarised in spending power.
2/ Where do the new jobs come from?
Brett Winton’s case is built on an innovation optimist’s dream, where barriers to starting a businesses collapse. A competent person with AI tools can ship products that previously required a team of ten. There are unexploited opportunities sitting in big tech - monopolised talent, underutilised capabilities, products that should be better but aren’t. AI will set them free.
The problem, as Shah points out, is that these new ventures are staffed at a fraction of the staff incumbents employ per unit of revenue. The pie may grow but the labour-intensive slice shrinks.
Andrew Yang goes further, to the career ladder itself:
“All of the experience that maybe we got in our 20s and 30s, there might not be that kind of ladder available to them.”
Zoe Scaman, viewing all this from within our industry, comes armed with evidence. Workers aged 20 to 24 held 10.5% of advertising and PR jobs in 2019. By 2024 that number was 6.5%. Global job postings requiring zero to two years of experience have dropped an average of 29 percentage points since January 2024. The average new hire in the US is now 42-years-old.
These may not yet be redundancies. It’s quieter than that. A “low-firing, low-hiring” equilibrium where headcount reduction happens not through sacking people but through choosing not to replace them when they leave.
What this means for planning: Agency partners are going to get older, leaner and more expensive per head, while the bench of emerging talent gets hollowed out. And as Zoe Scaman says, no agency has any incentive to change this state of affairs.
3/ Complement or substitute?
If AI is turbo-charging experienced senior practitioners, that is a deflationary phenomenon. Better output for less labour input. At 33_Zero we’re delivering quant insights for clients who never previously had the resources to pay for research agencies when undertaking strategy work. That makes the work we do for them more effective and their outcomes better with barely any meaningful additional cost.
Satya Nadella extrapolates this to the enterprise-level efficiency: the IP is in orchestration and context engineering. The firm that feeds its unique data and tacit knowledge into AI models gains sovereignty over its own value.
But the substitution curve is steepening. For the first time, machine intelligence can become a direct substitute for human intelligence at one-hundredth or one-thousandth of the cost.
The Citrini scenario imagines a procurement manager negotiating a SaaS renewal by credibly threatening to replace the vendor entirely using AI tools. In their fictionalised account the company renews at a 30% discount.
“That was a good outcome,” the scenario notes. “The long tail of SaaS had it worse.”
In most organisations right now, AI is augmenting experienced people’s judgement while automating execution that junior teams once delivered. That’s the complement phase. The substitute phase comes when the augmented senior person realises they no longer need the team around them at all - just the tools.
What this means for planning: Map your marketing operations against this split. Client relationships, strategy, taste, judgement and creative direction sit on the augmentation side. Research synthesis, churn content , campaign reporting etc. sit on the substitution side. The substituted functions are where cost comes out. The augmented functions are where differentiation lives. If you’re still charging junior rates for execution that AI now handles, that’s a workforce structure already looking outdated.
4/ The pipeline problem
Zoe Scaman’s “The Pipeline Problem” piece deserves quoting here:
“Every senior operator celebrating their AI-augmented productivity got there through a process that is now being dismantled... I had mentors who gave a shit. Job security that let me experiment. Room to be wrong without it being career-ending. The judgment I now leverage - knowing what good looks like before it’s made, reading a client’s silence when the work is in trouble - didn’t arrive with the senior title. It was built. Slowly. Through years of doing things badly and being told why.”
“The industry is waving this model around with jazz hands while quietly gutting the very conditions that produced it. It’s using the last generation of properly developed talent to argue that properly developing talent is no longer necessary.”
Remember, the average new advertising hire in the US is now 42 years old.
She gives complacent holding companies a sharp jab in the ribs:
“The industry didn’t start dismantling this pipeline with AI. Training budgets were the first casualty of every downturn for twenty years. Graduate schemes were hollowed out long before anyone had heard of a large language model. AI just gave the industry a respectable rationale for what it was already doing.”
She uses a football analogy. Clubs invest millions in youth development knowing most of those players will never make the first team. But they do it because they understand something the creative industry hasn’t quite grasped: the health of the ecosystem you operate in is a precondition for your own success.
What this means for planning: Client-side teams with the right senior hires can increasingly do execution work themselves. That’s the short-term opportunity. The medium-term risk is that the external talent pool you draw on for strategic thinking, creative direction and cultural fluency is drying up. The model works until the seniors retire, burn out or move on, and there’s nobody behind them.
5/ Is the cost curve as smooth as advertised?
The way Satya Nadella frames it sounds great: tokens per dollar per watt equals GDP growth. Token pricing drops by half every three months. Cheaper commodity, better growth.
Ed Zitron’s reporting complicates this picture:
“The entire AI infrastructure buildout depends on debt - venture capital, private credit, bank lending - at a time when global financial plumbing is under stress. Lifetime revenues are materially lower than the sum of reported annualised figures, which means the growth narrative underpinning hundreds of billions in infrastructure investment is softer than the headlines suggest.”
Meanwhile, the Iran crisis has driven energy costs up sharply. Natural gas powers the data centres that power the models that power the tokens.
None of this means AI capabilities stop advancing. But it does mean the smooth, monotonic cost decline that underpins corporate planning may prove lumpier than expected. If energy costs spike, if debt markets tighten, if the capex cycle stutters, the assumption that everything just keeps getting cheaper needs testing.
What this means for planning: Don’t anchor AI cost projections to the trajectory of the last eighteen months. Build in scenarios where capability keeps improving but the cost curve flattens or temporarily reverses. The technology can do more and more. But the financial infrastructure supporting it looks undeniably fragile. We may look back on this AI onboarding period as a one-off bargain deal to attract new users, never to be repeated.
What this means for the year ahead
The optimists might be right. The pessimists might be right. What seems sensible is to build some optionality for both:
Think carefully about how consumer spending in your categories might change if the middle of the market gets squeezed. Look honestly at the complement/substitute split in your own operations and in the agencies and partners you work with. Consider building in-house capability for the execution work that AI handles well, while protecting access to the strategic talent that’s getting scarcer. And stress-test the cost assumptions in any plan that depends on AI infrastructure economics running smoothly.
The debate is going to keep running. The planning really can’t wait for it to resolve.
Sources:
About 33_Zero
33_Zero works with brands large (AWS, Oxfam) and small (Agronomics, Ivy Farm) on brand and comms. Our clients recognise that unprecedented change needn’t be a threat but an opportunity. We help your brand show up and participate in this new reality.
Email jamesp@33seconds.co or subscribe and DM us here to find out more.