The AI Taste Gap: Why most AI transformation programs are solving the wrong problem

I lead R/GA’s enablement program. Everything that follows comes from doing this work across creative, strategy, technology, client engagement, delivery management, finance, people and operations over the past twelve months. These are observations from practice.

Most businesses are investing heavily in AI to do their current work faster. They’ll succeed. And it won’t matter.

When every business achieves the same efficiency gains at roughly the same time, the advantage disappears. A deck that took eight hours now takes two. A research sprint that took three weeks now takes three days. If everybody else has done the same you haven’t gained a competitive advantage, you’ve maintained the right to be at the table. You’ve built a faster horse.

We were seeing this as early as August of last year. [This Digiday article](https://digiday.com/marketing/as-ai-alters-cost-of-creative-indie-agencies-review-how-they-charge-clients/) outlined that as agencies adopt AI to recover years of margin erosion, clients respond by expecting those savings passed through as lower fees. 

The uncomfortable truth is that the vast majority of AI transformation programs are optimising for the wrong thing. They’re focused on speed, cost reduction, and automation of existing workflows. All of which are necessary. None of which are sufficient. The real question is what happens after automation, when machines handle everything that can be specified. What’s left? And is your business investing in it?

The shift happening right now is bigger than most people realise, and the gap between where businesses are and where they need to be is measured in orders of magnitude.

[Azeem Azhar’s “Magnitudes of Intelligence” framework](https://www.exponentialview.co/p/the-hundred-million-token-day) maps AI usage using the Eames’ Powers of Ten as a lens:

- At a thousand tokens a day, AI is a tool. Useful but bounded. A question here, a rewrite there. 
- At a million tokens, AI is a workflow. Embedded in daily process, shaping how work gets done. 
- At a hundred million, AI is a workforce. Autonomous agents that operate overnight, produce the equivalent of days of human output, and report back in the morning. 

Azhar personally reached a hundred million tokens a day in February 2026. Five orders of magnitude in thirty-nine months.

Most organisations currently sit at the tool level. The more ambitious ones are building programs to reach the workflow level. Almost nobody is planning for the workforce level, where agents run autonomously and the nature of the work itself changes.

That work was necessary. But it was also, in hindsight, table stakes. The competitive advantage lives somewhere else entirely.

Here’s the trap most businesses fall into. They automate existing workflows. They measure success by time saved. They celebrate when a deliverable that used to take a week now takes a day. And then they wonder why their margins haven’t improved.

The reason is simple. In a competitive market, efficiency gains compress margin. If you can produce a strategy deck in two hours, so can every other firm using the same tools. Clients will expect that speed as standard and push prices down accordingly. You’ve reduced your cost base, but you’ve also reduced what the market will pay for the output.

[Ethan Mollick’s research at the Wharton School](https://www.oneusefulthing.org/p/management-as-ai-superpower) illuminates why this happens. He frames AI delegation as an equation: human baseline time versus probability of AI success versus the time to request, wait, and evaluate. The equation works when you know what good looks like and can specify it clearly. Most programs focus obsessively on the tools, the efficiency, the AI process time. Far fewer invest in developing the judgment to specify well, evaluate accurately, and know when the output is genuinely good versus merely plausible. That judgment is where the value lives. And it’s the part that can’t be automated.

When machines handle everything that can be specified, value migrates to what resists specification. That’s taste. Domain expertise applied as quality judgment. The ability to know what should exist but doesn’t yet.

Running an AI transformation program has taught me that four capabilities consistently resist automation, regardless of department or business unit. They emerged as patterns across every function we worked with, and they’re where human value concentrates as AI takes on more of the execution:

**Curiosity and asking the right questions.** This is the upstream taste function and the most important one. Every AI output is only as good as the question that prompted it. The most expensive AI mistake is a well-executed answer to a question nobody should have asked. The ability to interrogate a brief, challenge an assumption, reframe a problem entirely. That requires knowing what you don’t know, and no model has that capability.

**Expert judgment applied with real-world context.** Can you look at an AI’s output and know whether it’s right for this client, this market, this moment? AI can generate a hundred options. Knowing which one is right requires years of domain expertise combined with context that lives outside any system: a competitor’s misstep last week, a regulatory shift that hasn’t hit the news yet, the CEO’s unspoken anxiety about the board. Learned expertise and lived context, fused into judgment. That combination can’t be trained into a model because half of it changes daily and the other half took a career to build.

**Narrative synthesis.** Tying everything together into a story that makes sense in the real world. AI can produce the components. Humans weave them into something that moves people and moves business. The difference between a collection of accurate outputs and a compelling argument is the ability to see the thread that connects them and pull it tight.

**Human relationship.** Successful service businesses are ultimately built on two things: strength of relationship and quality of output. Both drive the metrics clients actually care about. No agent replaces the trust built over years of working together, the ability to read a room, to know when to push a client and when to hold back, to have the difficult conversation that protects them from a bad decision. The human-to-human connection is where loyalty, retention, and long-term commercial value live.

There’s a shift happening in how value gets created. Historically, businesses like ours sold the things we made: the strategy, the campaign, the experience. Increasingly, the real product is the system that produces those things. The prompts that encode our best strategic thinking. 

The agent that holds everything we know about a client’s brand, market, and preferences. The workflow that turns a brief into a tested prototype in hours. The scaffolding that lets AI do the production work while humans focus on the judgment calls. We’ve gone from being a business that makes things to a business that makes things that make things. And the winners will be the ones who combine taste and tools most effectively: deep domain judgment directing powerful AI systems toward the right problems, producing the right outputs, for the right reasons. Tools without taste produces volume. Taste without tools is too slow. The combination is the competitive advantage.

[Mollick’s research reinforces this.](https://www.oneusefulthing.org/p/management-as-ai-superpower) His MBA students built viable startups in four days using AI. 

They succeeded because they knew how to scope problems, define deliverables, and recognise quality. The skill that mattered was management: the ability to specify, delegate, and judge. Think of it like an elite cyclist. They don’t just ride the bike. They can change any part of it. That’s the level of fluency every knowledge worker needs: deep enough to tinker with and fix an agent, while delegating the bulk of execution. Value is shifting from “I can do the work” to “I can specify, delegate, and judge the work.” That applies to a junior strategist as much as a creative director, to a finance analyst as much as a project manager. Across every function and every level, the people who can combine domain expertise with the ability to orchestrate AI will be the ones whose contribution is hardest to replace.

If taste is where value lives, the strategic question becomes: are you developing it as an organisational capability, or just hoping your best people have enough of it?

Practically, businesses need to operate on two time horizons simultaneously. Most are only investing seriously in one.

**Horizon 1 is foundations.** Fix tool access. Write the policies. Drive adoption. Automate current workflows. This is necessary, unglamorous, and competitively essential. Every serious competitor will reach this level within six to twelve months. Do it well, but understand what it is: the price of entry.

Be honest about how hard even this is. Adoption in particular is brutal. You can provide the best tools and the most inspiring training. Then Monday morning arrives, deadlines are tight, and fifteen years of muscle memory wins. The businesses that crack Horizon 1 treat adoption as a behaviour change problem, and engineer the environment accordingly: mandatory protected time for AI development, tools embedded at the point of need in existing workflows, AI skills written into career progression, and hiring AI-native junior talent who normalise new ways of working for everyone around them.

**Horizon 2 is where the real transformation lives.** Reimagine workflows from first principles. Build agentic infrastructure: persistent client knowledge agents, automated pricing, systems that work overnight and report back. Develop taste as an organisational capability. Start producing fundamentally different outputs: interactive artifacts instead of decks, experiences instead of documents, things that show rather than tell. Invest in the management and delegation skills your people need to orchestrate AI.

This is where the commercial model genuinely shifts, because you’re selling judgment, relationships, and the systems that make AI work for a specific client’s context.

The businesses that only invest in Horizon 1 will be efficient and undifferentiated. The ones that invest in both will be positioned to create and capture genuinely new value. And the ones who lack the ambition to push past Horizon 1, who nod along when you describe Horizon 2 but go back to optimising what they already have, will discover that efficiency without differentiation is a slow way to become irrelevant.

The taste gap is an ambition gap. The question is whether you’ll close it.

Tools are abundant. Taste is scarce. Which are you investing in?


tastea-ai-editorial-1

Subscribe below to discover how businesses can get ahead in the Intelligence Age.

R/GA, Marketing Agency

The AI Taste Gap: Why most AI transformation programs are solving the wrong problem