Five AI Debates Every Board Is Having (and How I'd Settle Them)

I'm the CEO of Superhuman and an active board member at several companies, including Spotify, Walmart, Schoolhouse.world, and, of course, Superhuman. This means I spend a lot of time talking with boards about AI. A handful of the same debates come up again and again. Here are the four questions I hear most often and how I think about each one.

The most common boardroom debate right now is: do we need fewer people now that we have AI?

There are clearly places where AI is making execution more efficient. But I think the net impact of AI will be more jobs, not fewer. The idea that jobs will simply be replaced assumes the amount of work in the world is fixed. History suggests the opposite. When power tools arrived, people worried they would replace carpenters. But who could have imagined, at the time, that we would build skyscrapers? New tools do not just reduce labor; they expand what we can imagine building.

That’s Jevons’ paradox: when a resource becomes easier or cheaper to use, total consumption often rises. Coal use did not fall when coal-burning engines became more efficient; it soared, because coal could suddenly power things no one had previously imagined. We are seeing a version of this at Superhuman, specifically on our Support Team. Our custom-built support agents now handle more than 72% of common support requests, but our ticket volume has increased by over 60% year over year, proving that better support efficiency expands demand rather than constraining it.

I had an interesting conversation about this with Bret Taylor, CEO of Sierra. His point was simple: give two companies the same tools, and if one uses them to build new things while the other uses them to do the same work with fewer people, who wins? In my view, it’s the company that dreams bigger.

Some jobs will clearly become unnecessary. But when a technology as big as AI arrives, it also creates opportunities no one could have planned for in advance.

In Age of Revolutions, Fareed Zakaria writes about many of the disruptions of the late 19th century: the railroad, electricity, the telegraph. The railroad transformed how goods and people moved—but did we still need wagon drivers? Electricity lit up cities—but did we still need lamplighters?

That framing is another version of zero-sum thinking. Each disruption also opened whole economies no one could have anticipated. The telegraph, for instance, was the predecessor to the internet—which gave rise to millions of jobs no one had words for yet. Today, 23% of 12- to 15-year-olds want to be content creators, a career that did not exist 30 years ago.

Across all of these disruptions, I see a consistent pattern. The jobs that disappear are visible and immediate, and the ones that emerge are not yet imaginable. But for the person living through the transition, the impact of their job disappearing hits right away, while the idea of new jobs appearing feels like an abstract future. Bridging that gap is a real challenge, and it requires action.

Walmart is a fantastic example here. When I joined their board, I had a sense of its scale. It's a Fortune 1 company with more than $700 billion in revenue and extraordinary reach across communities nationwide. But I was amazed to learn about Walmart Academy, the nation's largest private training program. Last year alone, 300,000 associates went through Academy training. Walmart is creating pathways into higher-paying, in-demand roles such as store management, truck driving, pharmacy, HVAC, and more.

That gap—between the job that disappeared and the one that does not yet have a name—is where boards can make a difference. They don’t need to predict the future, but they can encourage companies to invest in the bridge.

A common question at every company is "how many tokens are we using?" It may seem like a concrete way to measure AI adoption, but it is mostly a vanity metric.

Think of horsepower. To this day, we still rate engines by how many horses they could replace. This unit made sense when the world was moving from horses to engines, but it's increasingly detached from what engines actually do. I think of counting tokens as the horsepower of AI. It's a way of keeping score that will look dated fast.

So what should you measure instead?

We use several metrics at Superhuman, but one that has been particularly powerful is borrowed from engineering industry best practices: true throughput. It measures a complexity-weighted score of code contributions per week, giving more weight to more complex changes than to trivial ones. By that measure, our output per engineer has more than doubled compared to the same period last year. We're using tokens to get there, but tokens are not the point.

In every boardroom, for every discipline, the question should be: what is the right measure? True throughput works for engineering. Other functions, like sales, marketing, finance, support, and operations, are all being transformed by AI in different ways, and each one needs its own metric. The specific measure will look different, but the principle will remain the same: stop counting activity, and start measuring whether the work itself is accelerating.

If you ask a company which AI tools they use, you're likely to hear about the newest model, maybe an up-and-coming product a vendor demoed last quarter. But the AI they actually depend on tends to be the stuff they don’t think of as AI.

Grammarly, one of Superhuman’s flagship products, is a good example. It fixes sentences as you write, without asking you to open a separate tool or prompt a model. It shows up inside the work you’re already doing. And even though Grammarly makes more than 100 billion LLM calls per week, most people don’t think, “I’m using AI right now” when they accept one of its suggestions.

That reminds me of something my dad said when I was in college. I had been offered a research job at MIT’s AI lab, and I called him to ask what he thought. He immediately said, “Why would you do that? AI is where all good technology goes to die.” His point was simple: “The moment something works, we stop calling it AI.”

That still feels true. The best AI often comes to you. It meets you where you already are, in the flow of the work you are already doing. So when a board asks if they are behind on AI, I tell them to look beyond the frontier models and big names. AI is probably already running in the business, and they are just calling it something else.

If you look back at these debates, you'll notice that they all stem from the same worry: what is AI going to take away? Does AI equal fewer people? Will roles disappear? But conversations about AI don't need to start from a place of anxiety.

The companies I see pulling ahead are asking the opposite questions. They don't want to know what AI will let them cut, but what it will let their people build. Before your next AI conversation with your board, find out where AI is already running in your business, even if it's being called something else. Worry less about usage numbers and start looking at whether the work is measurably getting better. And when someone suggests trimming headcount, try to think instead about what new things you'd be able to build if you didn't.

The definition of AI will keep moving. The winners won't be the companies that used it to do the same work with less. They'll be the ones who used it to do something they couldn't before.