The critical skill workers need in an AI-powered world

Workplace AI expert and author Matt Beane explains what interactional expertise is and why your career may depend on it

Howard Rabinowitz

Howard RabinowitzThe Works contributor

Nov 04, 20244 MINS READ

Communication, leadership, and problem-solving are among the most in-demand skills in today’s talent marketplace, according to LinkedIn’s 2024 Most In-Demand Skills list. No surprise there—they’re all well-established competencies. But as employees begin to rely more on AI to get work done—and as new forms of AI, such as AI agents, become more like colleagues than just bots—does that change the calculus for the ideal skill set of the future?

The answer is yes, says Matt Beane, a professor of technology management at UC Santa Barbara and author of the new book “The Skill Code: How to Save Human Ability in an Age of Intelligent Machines.” In the book, Beane challenges conventional wisdom about workplace skills and identifies what he considers a new meta-skill for employees: “interactional expertise.”

As Beane argues, interactional expertise is the ability to tap AI’s vast knowledge base and quickly become conversant in a new function of business. In a recent interview with The Works, Beane explained why this ability is so valuable and how employers can start to recruit for it.

What is interactional expertise, and why do you think it’s so essential?

That’s the ability to discuss skilled work fluently without being able to perform it. 

Last year I taught a group of master’s students with no coding expertise. During a quarter-long intensive coding program, they built a functional web-based software tool for project managers. Did that mean they were skilled coders? Clearly no, but as one of my students put it, cynically but accurately, “I think I could gaslight an employer into hiring me as a software engineer.”

She didn’t build coding skills, but she had built interactional expertise. She could use technology to be conversant in a territory that she didn’t have deep practical experience in. That allows you to collaborate across functions, disciplines, occupations, or organizational boundaries—with coworkers, vendors, and customers, and without a degree or a certification. 

Interactional expertise doesn’t just mean you’re able to collaborate with human coworkers. It also allows you to work well with AI agents.

Matt Beane

Professor of technology management, UC Santa Barbara

You’re not an expert in their area, but you can talk with them about their work in a way where they see, “Oh, you get the basics.” That’s the new standard. It means you can be part of a business in a practical way that wasn’t possible before.

Anybody who says they know one particular skill that you need to learn in order to function well in an AI-suffused world is deeply deluded. Things are changing way too fast for that. If you spent money on a course on prompt engineering last year, that was worth it for maybe six months. Now systems are going to handle prompting in a way that makes that skill far less relevant. But interactional expertise? That’s going to become exponentially more important as we collaborate more and more across boundaries.

How does this ability translate into business value?

I defy anyone to look at complex, valuable work and not see the need for interactional expertise. Whether it’s software development, expanding into a new market, or navigating a complicated merger, the lack of an ability to coordinate fluidly can be costly in terms of time and money. A key part of these costs has to do with understanding your counterparts’ reality well enough to give them useful inputs and handle their outputs well.

Take software development. You’re a marketing manager working with a senior developer facing a tough roadblock. You don’t have skills in coding in Python or understand software infra or QA, but you can use a tool like GPT4+ to get up to speed and interact with that developer in a way where you can talk with them about their work.

Interactional expertise doesn’t just mean you’re able to collaborate with human coworkers. It also allows you to work well with AI agents, which is starting to happen and will only become more critical. 

How should companies go about evaluating their employees and job candidates for interactional expertise? 

You want to test for how they approach becoming conversant in a territory that they have no practical experience in. Say a person is coming in for an HR role. Let them know you’d like to discuss a specific area outside their experience, such as business development strategy.

People-first AI is transforming service. Are you ready?

Freshworks AI Summit: Register now

Give them a few minutes to prepare using their phone any way they like. They can use Google if they want or, ideally, on their own initiative, a large language model. Can they fluidly discuss a new business arena with little warning or prompting? If they say, “Geez, I don’t know,” that tells you something. That means they’re not aggressively using the AI that’s available to them. 

How might interactional expertise be relevant for the agentic AI that’s emerging? 

With AI agents, we’re poised to work with intelligent machines in ways that aren’t entirely clear at the moment. We’ll need to develop new skills to do that successfully. This is not an on-off switch that’s going to get flipped. This is a continuum and we’re going to see progress up close. Over the next six to 12 months, I think we’ll see a steady uptick in the length of time AI can work without human supervision or correction and the complexity of the problems that these systems can eat. 

Interactional expertise will be required to understand the inputs, processes, and outputs of these systems. Relatedly, specific skills that are liable to be needed fall into the managerial bucket. It’s very reasonable to say that if you can’t manage people, you’re going to have trouble managing agentic AI. But there will be different tactics you’ll need to deploy with agentic AI. As you delegate tasks to a non-human agent, it will mess up and you’ll have to correct it. You’re going to have to manage throughput and assess, “Is this good work?” Although we have a lot to learn there, it’s really exciting territory.