A Review of James Wang's "What You Need to Know About AI"

Can Shahrukh faithfully reconstruct the arguments of a book on AI written by a GP? And fairly critique its shortcomings?

Today’s post is an experimental one: reviewing a fund manager’s published work.

It is dedicated to meditations on James Wang’s new book, What You Need To Know About AI: A Primer on Being Human in an Artificially Intelligent World.

James, general partner at the venture capital firm Creative Ventures, has been a thoughtful reader of Cash and Carried.

We initially connected because he pointed out a boo-boo I made in an article on equalization interest. And here we are many months later: I have the absolute pleasure of reviewing his book, which comfortably took me about three days to read cover-to-cover.

It is out October 15^th, and you can order it here.

My main takeaway?

James thoughtfully accomplishes the goal he sets out at the beginning: of providing those without technical backgrounds the knowledge to adequately grasp the technologies behind and broader social questions around AI. The book is strong on fundamentals—though it occasionally reads as if written too deeply within the echo chamber of Silicon Valley’s techno-optimism.

Ultimately, James argues AI will augment our humanity; I push back in that it will reveal how much of that humanity depends on physical friction and the inefficiencies we refuse to automate.

As such, this post has three parts:

Part I introduces my reservations about James’ arguments as framed within a personal accounting about the enduring value of physical labor.

Parts II-A and II-B offer a more objective recounting of the book’s contents.

Part III rounds out a completion of scattered, but critical observations.

---

Part I: Within the API

I’ll start with an anecdote, in which the human operates in conjunction with the machine—without which the machine itself is of limited value.

It’s 7:30am on a crisp September afternoon, and two men dressed in matching checkered blazers, tucked-in button downs, and jeans (co-founding tech bros on a roadshow looking to IPO their Series G startup) step out of the Marriott Marquis in Times Square.

The day’s ride: an Escalade ESV Sport.

Blacked out alloy rims with a polished finish. Air ride adaptive suspension. The best electronic limited-slip differential money can buy. A massive dash display. And fully tinted windows, possible only because of a permit from the state DMV.

Cadillac, as expected, spared nothing in its latest iteration.

And I guess that’s why the Escalade continues to be beloved by those who can afford it. The physicality of it remains intimidating. Bloated. Majestic even.

Waiting by the curb, smiling, is my older cousin—a man who’s been a limo driver for two decades, and the first person to ever tell me what private equity was.

His passengers have included executives, royalty, and athletes. When he was allowed to share stories, they were always worth hearing.

Because of all the change that companies like have Uber propagated, industry analysts have suggested that those like my cousin (he doesn’t work with Uber) live “below the API.”

The argument is as follows: modern gig platforms replace middle management with software that automates human coordination.

This “software layer” includes a user interface for customers (created by those who sit “above the API”), the programming interface (API) that dispatches workers, and a worker interface for task execution. The API computerizes decisions once made by managers.

For example, Uber’s system takes payment and trip details, then dispatches a driver. This efficiency, it has been argued, widens the gap between automated work and high-leverage, technology-driven roles.

I find this proposition to be thoroughly inaccurate.

Certain types of high-end service providers (like corporate lawyers or bankers) do indeed get paid for their expertise. Others get paid for their physical labor, the delivery of which is automated by programmers and corporates above them. But it’s more than that. My cousin’s success requires emotional acumen, intense attention to detail, and equal amounts of professionalism.

If anything, he lives not above or below, but in the hinterlands of the API: he receives instructions from a computer, but his execution requires extreme discretion and discernment, none of which can be assuaged by technology. Rich people are, after all, picky and petty.

Empire CLS, Blacklane, and the ultra high-end bespoke transportation provider Fortis—based out of the innocuously unassuming city of Greenville, South Carolina—are among such companies that have human dispatchers and hand-picked chauffeurs who go through rigorous, reference-based background checks.

Some jobs are menial, but they require mastery of a certain kind. Uber cannot dispatch a fully-loaded S-Class to ferry a defected oligarch’s son to a rugby match 5 minutes away. The average Uber driver goes from point A to point B. But that’s too formulaic for some people.

Because we want the damn ritual functions too. The opening of the car door. The smile. The passersby watching as the door is opened for us again!

What drives this type of consumption in a way that escapes the automation-first principles that drive the hype around AI?

While those like Fortis technologized to some degree (phones, cybersecurity, the latest cars, etc.), their business model never betrayed the long-standing belief that there was always a class of consumer for whom automation was a dirty word. And this class of consumer was always willing to pay a premium for the elements of service that constituted the absolute antithesis of automation.

We may simply describe this antithesis as “the human touch.”

If my cousin’s world proves anything, it’s that certain forms of value exist outside the cachet of AI precisely because they resist programming code-based abstraction.

Part II-A: The History of AI

I might be reading too much into it, but James unintentionally admits as much about the importance of the physical: the bookmark that came with the book is soft to the touch, its iridescent ink catching light at different angles.

The object insists on being noticed by a weirdo like me (can physical objects ever insist on anything?).

Still, I might even laminate the bookmark for nostalgic purposes!

My copy and the reflective bookmark!!

It was also heartwarming when I saw a handwritten note from James on the first page (how would I have felt if it were AI-generated?).

The book’s early chapters trace AI’s technical and historical foundations before shifting toward its economic and moral frontiers. Here, James is at his best: fluent in both the engineering and the incentives that have driven AI.

He notably describes AI as “more alien than alien,” noting that even extraterrestrial life would be bound by drives for self-preservation, nourishment, and reproduction, whereas AI requires none of these and can be “propagated endlessly by its creators.”

Two schools of thought emerged in the 1950s, each fueling a distinct boom cycle: one in the 1960s and another in the 1980s.

The first, symbolic AI, centered on the explicit representation of knowledge and the logical rules governing its use. This approach was intentionally designed and reached its peak in the era of expert systems.

The second cycle was inspired by neuroscience and is familiar to us today: connectionism, which serves as the basis of neural networks that power LLMs. This philosophy relies on the distribution of “weights” across interconnected nodes, where learning emerges not from explicit rules (as with symbolic AI, and beyond which it cannot operate) but from the gradual adjustment of these weights in response to data, mirroring, in abstraction, how synaptic strengths evolve in the human brain.

Beneath the enchantment of AI’s human-like fluency, the book continues, is a mechanical thingamajig. That is, LLMs are “huge statistical machines” that simply autocomplete text. This text includes their own output, until some kind of coherence emerges.

Further, LLMs are not self-aware, only probabilistic. They “wander” through space to find the most fitting continuation to a prompt, producing the illusion of reasoning.

He is right to demystify LLMs as probabilistic engines, though his framing risks overstating their harmlessness. Because automation has social costs beyond misprediction.

This stuff many of us know very well, whether from popular commentary or general conversations with friends and colleagues.

Part II-B: The Economics of AI

The book’s most original insight concerns data. Specifically, the tension between the frictionless digital world, where information flows habitually, and the physical world, where data is costly, unreliable, and scarce.

That resistance, James argues, is where defensibility lies: even as computation becomes commoditized, competitive advantage rests on hard-won, proprietary datasets others cannot easily replicate.

He sketches three archetypes of AI enterprise—data monopolies, cost monopolies, and product innovators. But the force of his argument lies in how he links scarcity to value. No matter how powerful the system, Amdahl’s Law reminds us that performance remains bounded by what cannot be parallelized—or, in economic terms, by what remains stubbornly human.

For me, though, the most illuminating chapter was James’ explanation of how GPUs actually work. He unpacks what made NVIDIA, the chief beneficiary of the GPU revolution, so extraordinarily successful: a blend of early experimentation in parallel processing hardware and Jensen Huang’s remarkable commercial resilience during the lean Obama years.

Various of my stickies with random reflections. Yes, I still reference Hegel and Mill.

Later chapters turn philosophical. James reflects on the “holiness of the physical,” suggesting that in an age of digital abundance, tactile technologies may regain sacred status. He closes with a look toward robotics, where AI seeks to generalize dexterity as language models have done with words—but where the friction of the real world still resists capture.

Part III: The Body and The Cloud

Still, the book’s optimism invites a practical question. What happens to those whose work is neither scalable nor sacred? Efficiency is seductive, but it is also a social experiment few have consented to—a process not unfamiliar to people working in legacy industries.

My chief rejoinder? James is not wrong about AI’s promise, but he does understate our attachment to the tangible. Some forms of value endure precisely because they resist automation.

To continue the anecdote from above, my cousin’s Escalade is hand-washed daily. He stocks bottled Fiji water, several phone chargers, and sanitizing wipes. The cabin is scent-neutral, and his English is impeccable.

He further needs to be socially agile. Some passengers welcome conversation. James himself, as he notes in his book, certainly does, seeing chatting as a way to take the pulse of a city while traveling (self-driving cars like the ones Waymo produces could never compete with the authenticity of butterfingered broadsides from an Uber driver).

Others value silence. Some want a specific route to the airport, even if it is longer.

And so, my counter rests on a pretty simple conviction: the technologist may speak of transcendence, but his appetite betrays him. He wants the tangible and the bespoke. The anti-AI configuration that is extremely difficult to scale or automate.

Think about it. Handwritten letters, home cooked (or gourmet) meals, bespoke tailoring, concierge travel services, personal trainers, college admissions consultants, autographed Ali boxing gloves, a Picasso in your living room, afternoon tea with a princess.

Or, for yours truly, front row seats to a summer polo match in the Karakoram.

The physical world is where friction reasserts itself. The irony, of course, is that the more we automate, the more sacred the un-automated becomes.

More troublingly, the equation may be inverted: AI makes the world easier for humans, but, as David Temkin, Editor-in-Chief of In Formation magazine, observes, “Every day, computers are making people easier to use.”

One piece from the issue, “My Toyota Ratted Me Out,” by Rob Leathern of Hawkview Labs, captures this inversion vividly. Leathern discovered that his car’s analytics provider knew precisely how often he accelerated or braked at more than six miles per hour per second, and when his antilock brakes were triggered. He only learned who was collecting this data after receiving an email from an insurer congratulating him on being a “safe driver” (getting in touch with the insurer was a headache).

I won’t dwell on privacy—it deserves its own treatment—but its near-absence from the book’s discussion of AI is conspicuous. It merits as much attention as any other topic, yet is scarcely acknowledged. For someone who so clearly understands data, James seems to leave unattended important ethical considerations: surveillance, consent, and information asymmetry, for example.

I wouldn’t proffer this point as a critique of James’ normative position about the importance of human flourishing or about the issues that afflict data privacy.

Read less charitably, though, AI optimism can look like a plea for transcendence. A secular faith of sorts that frictionless systems will redeem us from our own messiness. But the opposite feels truer…because the mere facticity of messiness legitimates us as distinct from machines, no?

The book, to its credit, illuminates the paradox of our desire for physical luxury even as it glosses over its full implications. James’ faith in AI’s promise to expand human potential invites admiration—and argument.

What You Need to Know About AI deserves credit for provoking the conversation. But what lingers is the question of frictions AI can never erase. And, in a way, frictions we have never wanted erased to begin with.