The 56% Premium: AI Skills as the New MBA

Everyone’s calling AI skills the new MBA. They’re not wrong. But they’re not thinking far enough back in the stack.

PwC’s 2025 Global AI Jobs Barometer put a number on what a lot of founders have been feeling anecdotally: workers with advanced AI skills are commanding a 56% wage premium, double what it was just the year before. Job postings requiring AI skills grew 7.5% while total job postings fell 11.3%. In industries most exposed to AI, revenue per employee grew at 3x the rate of industries that aren’t. The signal is unambiguous.

So yes, AI skills are the new MBA. The credential that once separated the generalist from the strategist is being repriced in real time. But here’s the problem with stopping at that framing: it treats AI fluency as a single, stackable skill to be acquired, the way you’d acquire Excel or Salesforce. And that’s precisely the wrong mental model for founders trying to build durable competitive advantage.

What are AI skills, really?

Before you can build a talent strategy around AI skills, it helps to be precise about what you’re actually looking for, because the market isn’t.

Most job postings that list AI skills are measuring tool familiarity: Can this person use ChatGPT? Have they touched a RAG pipeline? Do they know what an agent is? That’s the floor, not the ceiling.

The skills that drive the wage premium are upstream of the tools. They’re the ability to evaluate what AI produces, to identify where in a workflow AI should and shouldn’t be applied, and to design information systems that AI can actually operate on reliably. In short: AI literacy, not just AI usage.

The WEF and Oxford Internet Institute found that in the UK, AI skills now command a higher wage premium than a master’s degree. But more pointedly, their research showed that demand for AI roles grew 21% as a share of all job vacancies from 2018 to 2023, while university education requirements for those same roles fell 15%. Employers are not hiring credentials. They’re hiring demonstrated capability to work with information at scale. That’s an important distinction, and it points directly to what the most in-demand AI skills in 2026 actually are.

The most in-demand AI skills in 2026 (and why one discipline teaches them best)

Here’s where the MBA framing starts to break down. The AI skills commanding a premium right now, things like evaluating sources, structuring knowledge, designing retrieval systems, and auditing outputs for quality and bias, aren’t new. They’ve been the core curriculum of information science programs for decades.

An MLIS degree (Master of Library and Information Science) is not the obvious answer when someone asks how to build AI competency on their team. But it probably should be in the conversation. Here’s why.

Marcia Bates described the core knowledge-seeking behavior back in 1989: people don’t search linearly. They move through information in iterative loops she called “berrypicking,” gathering bits as they go, revising their mental model with each new piece, adjusting what they’re looking for based on what they find. The modern AI-fluent worker is doing institutional berrypicking at scale, deciding which sources to trust, how to structure retrieval, when to stop and act. The model has changed. The underlying cognitive architecture hasn’t.

This is why information science, not computer science or business school, is the hidden prerequisite for the kind of AI skills that actually command a premium. RAG architecture, agentic workflows, enterprise knowledge bases: these are not engineering problems. They’re information architecture problems. The engineering is usually the easy part.

What the research says about AI upskilling

McKinsey’s research on learning and development for the AI age is blunt about the gap between what organizations are doing and what they need to do. Most companies running AI upskilling programs teach tool usage: specific interfaces, specific prompts, specific workflows, rather than the underlying principles that would let employees transfer those skills as the tools change. Given that the skills sought in AI-exposed roles are changing 66% faster than in other roles (per the PwC data), training to the tool is a losing bet. You’re always a product cycle behind.

Effective AI upskilling looks different. It builds mental models, not muscle memory. It teaches people why retrieval works the way it does, not just how to write a better prompt. It develops judgment, about when to trust the model, when to check the output, and when to override entirely. That’s closer to what a good MLIS program teaches than anything in a standard AI bootcamp.

McKinsey’s work on AI readiness at scale identifies the organizational inflection point more precisely: companies that outperform on AI ROI aren’t the ones with the biggest models or the most compute. They’re the ones where mid-level employees can independently identify where AI should be applied, design the application, and evaluate whether it worked. That’s an AI literacy problem, not a technology problem.

Knowledge management AI: the infrastructure no one is talking about

APQC’s 2026 Knowledge Management Predictions make a point that founders consistently underweight: the companies pulling ahead on AI aren’t treating it as a standalone capability, they’re integrating it into their knowledge management infrastructure.

This reframes the question entirely. It’s not “do we have AI tools?” It’s “do we have a knowledge architecture that AI can actually operate on?” Bad metadata, inconsistent taxonomy, tribal knowledge that lives nowhere but in people’s heads, these aren’t quaint legacy problems. They’re the reason most knowledge management AI investments underperform. The ROI is in the data and the structure, not the model.

For founders, this means the highest-leverage AI investment often isn’t a new tool. It’s a knowledge audit: map what your organization actually knows, where it lives, who holds it, and what happens when that person leaves. AI can then operate on that infrastructure. Without it, you’re running a powerful engine on contaminated fuel.

The AI agent skills gap

As agentic AI becomes mainstream, systems that take sequences of actions rather than just responding to single prompts, a new tier of AI agent skills is emerging that most organizations haven’t started to hire for or develop.

Operating AI agents well requires something most training programs don’t touch: understanding how to define scope, set decision boundaries, audit agent behavior, and recognize when an agent is producing confident-sounding nonsense. Those are governance and information architecture skills. They’re also, again, what good information science training produces.

The Deloitte 2026 State of AI in the Enterprise (n=3,235 leaders, 24 countries) found only 21% of organizations have mature governance for agentic AI. Most lack formal governance frameworks for agent behavior. The gap isn’t model capability. It’s human judgment about how to work with these systems responsibly, and that judgment has to be trained, not assumed.

Is the AI skills wage premium a bubble?

Is the 56% premium durable, or is it a temporary bubble inflated by novelty?

The honest answer: some of it will compress. Prompt engineering as a discrete job title is already deflating, AI skills at that level are becoming table stakes. But the premium for upstream capabilities, knowing what question to ask, how to evaluate the answer, and how to design systems that produce reliable outputs, that’s not going away. If anything, it grows as AI capability improves and the bottleneck shifts entirely to human judgment.

The MBA premium compressed too, eventually. But the underlying skills, strategic thinking, financial fluency, and organizational design, remained valuable. The same logic applies here. The wage premium will redistribute, but it will stay with the people who understand information, not just the ones who can operate the tools.

How to learn AI skills that actually hold value

If you’re a founder or operator thinking about what AI skills to learn or how to develop AI literacy on your team, a few things worth sitting with:

• On hiring: The next time you’re evaluating a candidate for an AI-facing role, ask less about which tools they use and more about how they decide which information to trust. That’s the durable capability. A candidate who has studied information architecture, knowledge systems, or even holds an MLIS degree may be more valuable than someone who has completed an AI bootcamp.
• On AI upskilling programs: Build them around principles, not interfaces. Teach your team why retrieval works the way it does, not just how to write a better prompt. Develop genuine AI literacy, the ability to evaluate outputs, design information flows, and govern AI behavior, rather than tool proficiency that expires with the next product update.
• On knowledge infrastructure: Before you add another AI tool to your stack, audit what your knowledge architecture looks like. Knowledge management and AI are not separate investments, they’re the same investment. The organizations that understand that are the ones outperforming on AI ROI right now.

The 56% wage premium is real. But the people who will hold onto it longest aren’t the ones who learned the most AI tools. They’re the ones who understood information first.