When the Skeptic and the Optimist Agree, Start Worrying

I agree with Cal Newport about AI. If you know us both, that sentence should concern you.

Cal Newport built his career interrogating technology—digital minimalism, deep work, pushing back against the attention economy. I've spent mine implementing it: blockchain infrastructure, AI governance frameworks, quantum-resistant cryptography. We're supposed to occupy opposite corners of the ring.

Last week, I wrote that both sides of the AI jobs debate are wrong. AI is genuinely disruptive, and it's not eliminating every white-collar job. Then two things landed on my desk within 24 hours. Cal published "The Dark Side of the Jevons Paradox" and reached my conclusion through completely different reasoning. And Dario Amodei—the man who spent 2023 warning of an AI "white-collar bloodbath"—sat onstage next to Jamie Dimon at Davos and grabbed for the same economic principle.

When the technology skeptic, the finance pragmatist, and the reformed doomer all land in the same place using different maps, pay attention.

The Jevons Paradox: Efficiency Creates Demand

The Jevons Paradox is named for William Stanley Jevons, a 19th-century economist who noticed something counterintuitive about coal. When James Watt's steam engine made coal burning dramatically more efficient, Britain didn't use less coal. They used exponentially more. Cheaper, more efficient energy didn't reduce demand—it exploded the number of uses people found for it.

If one programmer can do the work of five using AI pair programming tools, companies won't fire 80% of their developers. They'll build five times more software. The bottleneck shifts from "can we afford to build this?" to "what should we build?" The resource becomes abundant, so the constraint moves to judgment.

I watched this exact pattern play out with cloud computing. When AWS made infrastructure cheap and elastic in 2006, CFOs didn't slash IT budgets. They greenlit projects that would've been economically impossible before. The companies that thrived weren't the ones with the leanest infrastructure teams—they were the ones whose teams knew what to provision and when to kill underperforming experiments.

Cal's Dark Side Is Real. It's Just Not Inevitable.

Cal's insight is that every efficiency gain carries a dark side. Steam engines brought unprecedented industrial capacity—and Victorian London's choking smog. Email made asynchronous communication frictionless—and created the always-on inbox that traps knowledge workers in reactive mode.

His argument: AI will make information work so cheap and abundant that we'll drown in it. More reports no one reads. More meetings no one needed. More analysis paralysis dressed up as data-driven decision-making. The dark side isn't the technology itself—it's our inability to resist using it for everything once the marginal cost approaches zero.

And he's right. I've seen this in client engagements already. A financial services firm I advised last quarter deployed an AI tool that cut investor report preparation time by 70%. Their reaction? They didn't reduce headcount. They started generating three times as many variants of the same report "because we can now." The tool created efficiency. Leadership created meaningless work to fill the space.

But here's where I part ways with Cal's framing: the dark side isn't a feature of the technology. It's a skill gap. And skill gaps close—for the people who bother closing them.

Not Everyone Drowned When Email Arrived

Email didn't ruin every knowledge worker equally. Some drowned in their inboxes, reflexively checking every notification, confusing responsiveness with productivity. Others built filters, set boundaries, learned when to batch communications and when to go dark.

Same disruption. Different outcomes. The dividing line was who developed the skill to manage the tool instead of letting the tool manage them.

I'm watching the same sorting happen with AI right now. Some teams use ChatGPT as a crutch, outsourcing thinking to the autocomplete oracle and losing the judgment muscle that comes from doing hard analysis yourself. Others use it as a sparring partner—stress-testing their logic, exploring edge cases, accelerating the iteration loop while keeping the final call human.

The technology is identical. The outcome depends entirely on whether you're building judgment or delegating it.

The Timing Is Suspiciously Convenient

Here's the uncomfortable part nobody's saying out loud: notice when Amodei and Sam Altman turned optimistic about jobs.

Amodei spent 2023 warning that AI would cause a "white-collar bloodbath." Then Anthropic started lining up a potential IPO. Suddenly the message shifted: AI will augment workers, create new roles, the Jevons Paradox means more jobs, not fewer.

Altman followed the same arc. Dire warnings about transformative AI risk, congressional testimony about the need for regulation, then—as OpenAI's valuation climbs toward $100 billion—a much sunnier outlook on labor markets.

I'm not saying they're lying. I'm saying when someone who sells the disruption tells you not to worry about the disruption, that's not analysis—it's a sales pitch. Investors don't buy stock in companies promising to eliminate their customers' jobs. The messaging adapts to the funding round.

So when they tell you your job survives, believe them. Then ask yourself the question they're not asking: am I building the judgment to handle what comes next, or waiting for someone to tell me it's safe?

The Question That Matters

Here's what I'm asking the partners and controllers and audit leads I work with:

Can you articulate the judgment call in your work that AI can't replicate?

Not "what tasks do you do"—tasks get automated. Not "what expertise do you have"—expertise gets encoded. The question is: where in your workflow does the messy, context-dependent, politically loaded, judgment call happen? The moment where two technically correct answers lead to different outcomes, and you have to make the call based on things that don't fit in a prompt.

For an auditor, it's not running the tests—it's knowing when the tests are asking the wrong question. For a CFO, it's not building the model—it's knowing which assumptions to stress-test when the board's risk appetite shifts midyear. For a controller, it's not closing the books—it's deciding how aggressively to interpret a new standard when the answer determines whether you hit earnings guidance.

AI makes the repeatable parts cheaper. That moves the value to the unrepeatable parts. If you can't name what's unrepeatable in your role, you're in trouble—not because AI will take your job next quarter, but because you're not building the skill that matters when efficiency becomes table stakes.

What to Do Monday Morning

1. Audit your last week of work. What did you do that required judgment versus execution? If the ratio is below 30%, you're building the wrong skill.

2. Ask your team: what would we build if capacity wasn't the constraint? The Jevons Paradox means you're about to find out. If the answer is "more of the same," you're missing the opportunity.

3. Set one AI boundary this week. Use it to accelerate something, then force yourself to do the next version manually. You're not rejecting the tool—you're making sure it's sharpening your judgment, not replacing it.

The disruption is real. The paradox is real. The dark side is real.

But nothing about this is inevitable. The people who thrive won't be the ones who resist the efficiency or blindly embrace it. They'll be the ones who recognize that cheaper execution moves the game to better judgment—and start building that muscle now.

What's the unrepeatable part of your job? If you can't answer that, the Jevons Paradox won't save you.