What AI Is Really Competing Against in Knowledge

The argument that AI can't replace knowledge workers rests on a premise worth examining: that human work represents some kind of high bar, a standard of quality and insight that machines struggle to reach. Daniel Miessler, who's spent 25 years embedded in companies from Apple to Fortune 10 energy firms, has a different view. The work AI needs to match isn't nearly as good as we pretend.

Miessler's argument, laid out in a 75-minute video for his Unsupervised Learning channel, comes from someone who's seen the inside of hundreds of companies. Not as a visiting speaker, but as a consultant fixing their security problems, which means seeing how they actually work when nobody's performing for investors.

The Baseline Nobody Talks About

Start with a basic question: In most companies, can leadership quickly list all the work being done, what it costs, and whether it's good? Miessler's answer, drawn from direct experience: No. Not even close.

"When a leader asks this question to their staff, to their company—'Hey, I need a list of all work being done, what everyone is doing, how much this work is costing'—do you know what the answer is?" Miessler asks. "The answer is: 'Well, we could take a look. That's a good question. If you would like, I could put together a project.'"

That project, he notes, takes months or years. Large companies often spend half a million dollars bringing in McKinsey to figure out what their own employees do all day.

This isn't the standard AI needs to beat. This is the actual state of most knowledge work: unclear processes, inconsistent outputs, meetings about meetings, and knowledge trapped in individual heads that walks out the door every time someone changes jobs.

Miessler describes what he calls "soup sandwich" companies—military slang for complete chaos. Most large organizations, he argues, share similar problems. Vision isn't clear. Standard operating procedures either don't exist, conflict with each other, or sit unread in some forgotten SharePoint folder. The same task, given to two different employees with identical training, produces wildly different results.

"If Sarah does it, yeah, I love when Sarah produces this report," he explains. "Jim over here looked at the same process, got the same training, had the same onboarding. His output is absolute garbage."

The $50 Trillion Question

Global compensation for knowledge workers runs about $50 trillion annually—$10 trillion in the US alone. That's not a figure that suggests some kind of optimized, high-performing system. It's spending on a scale that invites scrutiny.

Gallup found only 21% of workers globally are engaged. McKinsey reported in 2017 that only 4% of US work requires creativity at median human levels. These aren't fringe observations. They're well-documented patterns that, for some reason, we ignore when discussing whether AI can match human performance.

The quality bar isn't where most people imagine it sits. Miessler points to the cultural touchstones we've been creating for decades: "Office Space," "Severance," the entire genre of corporate satire. We've been documenting workplace dysfunction and soul-crushing processes for so long that it's become background noise.

"Why do you think people dread Monday so much? Why do you think people on Sunday night are filled with this pit of despair?" Miessler asks. Not because they're heading to deeply fulfilling work that makes full use of their capabilities.

What's Actually Being Defended

The counterargument—that AI is "just a text generator" without real understanding—comes from smart people. Miessler mentions Marcus Hutchins, a respected security researcher who's debated him on this. The technical objection has merit in narrow contexts. But it misses what knowledge work actually is in most companies.

Most knowledge work isn't creative problem-solving. It's following processes, or trying to. It's taking inputs and producing outputs according to rules that may or may not be written down. It's work that gets done inconsistently because the person doing it is on vacation, or didn't read the memo, or is one of the dozens of employees who never got good at their job but are hard to fire.

"We are holding this situation up as the standard that AI can never achieve," Miessler argues. "It's absolutely insane to me."

The risk he identifies isn't that the skeptics are technically wrong about AI's current limitations. It's that people who accept that narrative won't prepare for what's coming. They'll assume their expertise and experience create a moat that doesn't actually exist—not because they aren't skilled, but because companies don't know how to capture, standardize, or scale what makes them valuable.

The Question Nobody's Asking

Here's what makes this argument worth taking seriously, even if you disagree with parts of it: Miessler isn't claiming AI is perfect or that all human work is worthless. He's pointing out that the comparison isn't between AI and idealized knowledge work. It's between AI and the actual work happening in most companies right now.

That work includes projects that take three months longer than they should because nobody documented the process. Reports that vary wildly depending on who writes them. Meetings that exist only to explain things that should have been in the email. Knowledge that evaporates when employees leave.

If you've worked in a large organization for any length of time, none of this is surprising. The question Miessler raises is why we're so confident AI can't do better. The debate about AI replacing knowledge workers often assumes knowledge work functions at a level that, empirically, it often doesn't. Whether AI can match human creativity or insight might matter less than whether it can follow a process consistently, document what it does, and not call in sick on Monday.

—Bob Reynolds has covered technology transitions since the mainframe era. He's watching this one with particular attention to who benefits and who pays.