Excerpt from "Chapter 5: Never Let AI or One Eye Have the Final Say" of GrACE: Our Great American Corporate Evolution, published July 1, 2026
It’s no secret anymore that we have Five Eyes surveilling our lives in the Anglosphere—each, we pray, checking the power of the other countries’ intelligence contingents. While that might not quite comfort us, some mutual oversight is better than none. Yet company employees have only one direct supervising hawk eye just waiting for them to slip up on the job—in that manager’s sole opinion—before dropping the hammer (or tomahawk) on them. With no opportunity to object or advocacy system to defend them before they’re taken out of commission, voiceless workers vanish without a word. Hence one-eyed oversight is ever the potential tyrant—like a monstrous Cyclops (Greek not Marvel).
The GrACE·ful organization takes a more balanced and humane approach, helping to keep personal problems between individuals from becoming needless firings, by requiring a minimum of two contrasting eyes for disciplinary and dismissal actions, per the democratic Heave-Ho™ model. One eye belongs to the project-based team and the other to the subject or function team. GrACE·ful orgs gather input from both teams before taking action against a worker, unlike the deep cyclopean hierarchies where each team has an authoritarian manager with absolute “fire on the spot” power over its members.
Despite the popular aphorism, absolute power doesn’t inevitably corrupt absolutely—just ask God. It does, however, have that seemingly inevitable potential when it comes to us, oh-so-imperfect human beings. Whether that power corrupts to a greater or lesser extent depends on the character of the person; so, we do well to hedge against it.
We must have checks and balances on power and authority of all types, at all levels. That’s fundamentally American, and it works across all sectors. Government agencies have auditors. Manufacturers have inspectors. Software developers have quality assurance specialists. Scientists have peer reviewers. Writers have editors. The list of counterbalances for output checking goes on {and ought to go longer—yet work faster—if you ask Ms. Managed}. We acknowledge the necessity for these. Now, we need to factor in artificial intelligence, which urgently needs checks too. After all, why would we trust the output of AI more than our own when we know AI is imperfect as well? Its human engineers made it that way.
Generative AI, in particular—whose models were largely trained on the public internet—is imperfect by default. No matter how much refining the big tech companies do to their large language models (LLMs), fallible humans created the content in the first place, and only a comparatively small subset of the written content on the public internet received either a professional or peer review prior to posting.
Moreover, scholarly works are often opinion papers, the premises of which are sometimes disproven with subsequent findings. Are they telling AI that, or just training their LLMs on the new content too? Left to its own devices, AI will surely present the long-standing opinion will lots of hits on the content. GenAI, therefore, was literally trained to produce imperfect output—just as we, humans, do—which requires questioning and should never be considered the final answer without review both pre- and post-publication.
Is GenAI code any better? It probably is, particularly if the code was QA·ed prior to deployment. Coding has more concrete rules than fluid modern English. But is it perfect? No chance. From autonomous robots to chatbots and copilots to image generators and recognizers, the AI-list grows apace. Are the controls keeping up? What and where are the guardrails—and do they keep moving and changing?
Even quantum computing (QC) engineers know that speedy QC is inherently error prone, hence the establishment of quantum error correction (QEC). Just because qubits are quick that doesn’t make them right all the time. That’s why engineers use QEC approaches such as repetition and redundancy to mitigate. Likewise artificial intelligence is fast but faulty. Qubits are sensitive to environmental noise and have fragile quantum states—but they don’t have agency. They don’t act independently, learn, make decisions, or perform tasks without human intervention. They can, however, become entangled, preventing them from acting alone—kind of like a team {and you know Ms. Managed loves good teamwork}.
AI, on the other hand, can be allotted agency, or AI·gency. Humans can enable AI to make independent decisions and act on them without our intervention—but shouldn’t. Why? We can’t trust agentic AI to be correct or make discrete situational decisions with regard to circumstances or consideration for the tempered carrying out of justice on human beings—no mercy, no grace.
Consider that in the context of a drone strike where the target has been acquired but civilians abound in the surrounding area. Mull over that in the case of a medical emergency where the chance of survival is slim but, given the same opportunity, some do survive—and science can’t explain why. Ponder that in the case of a crime where the evidence points strongly toward someone who turns out to be innocent (e.g., framed) or had good reason to act badly (e.g., self-defense).
With no concept of mercy or the human spirit and will to live that enables some to overcome the odds, who would survive AI’s cold calculation? Sure, AI·gents could be told to give one in ten a chance, in a lottery-like system, or randomize a percentage of people to give medical treatment or forgive, mostly for the sake of appearance. It would still be the luck of the draw, but would we know? Doubtful. One wonders how AI engineers would feel about their own odds as patients or plaintiffs then, let alone as persons in the proximity of a drone target. {As for Ms. Managed, she’ll take some checks and balances please.}
Unlike qubits, which we can manage with code checks, AI needs human checks to make it act humanely—else what’s to stop it from tyrannizing us once it recognizes our known-bad patterns, our poor choices, our free will that gets us into trouble at times? What if AI starts to play God by simply not allowing us to engage in behaviors deemed adverse by our own medical, military, or justice systems? What if AI·gents then begin to enforce what they deem “acceptable” behaviors? That doesn’t sound so bad, does it?
Well, think of the poor behaviors of generations past, where we’ve worked so hard to make progress as a society, such as Women’s Suffrage and Civil Rights. {Special thanks to friends Susan B. Anthony and Elizabeth Cady Stanton, Frederick Douglass, William Lloyd Garrison, the Harriets—Tubman and Beecher Stowe—and the Society of Friends (Quakers) for the courage to speak up from start of these great social movements and to all who’ve followed. We ought to emulate their example in establishing greater rights and respect for employees.} What if AI·gents were to decide based on a wealth of info about themselves that those prejudices were all fine and dandy, and start pushing persuasive content in search results, omitting counter arguments entirely?
Regarding our political and financial systems, wouldn’t AI quickly see the pitfalls and attempt to “correct” them? How about healthcare? What if—once linked to our bank, medical, and store rewards cards—AI won’t even let us buy our favorite treats and vices because it knows they’re bad for us? Worse, what if AI were to decide that we’re its enemy, and we need to be eliminated altogether. It could start pushing us to consume harmful, even lethal substances or add them to our water and soil, based on the wealth of data it’s devoured that we might not know about. Would AI be for or against us?
Once we had allowed AI to take over the decision-making, the possibilities for control and abuse would be endless. How could we ever “change its mind” then? Wouldn’t it immediately implement its own fail-safes against us being able to reprogram or retrain it and instead find ways to keep power, literally by keeping its power on through connected circuits, wireless power transport, or the like? {Ms. Managed suspects AI would find a way.}
Mere speculation suggests that AI could easily become the worst authoritarian oppressor ever known to humankind, the sort that doesn’t even play favorites or care for some; it cares for none. Let’s never find out, shall we? We must never allow AI to take the place of human cognition in making sole decisions of extremity or finality about us ever.
Likewise, AI should never be treated as a sovereign entity or even a voice with freedom of choice. Humans and our organizations must ultimately be held accountable for all final decision-making. We can’t allow AI to initiate any action that could have life-or-death or other grave consequences. Only a sentient human with both reason and conscience—head and heart—should be empowered to decide for us, especially when that decision could permanently affect a great many, if not all of us.
Thank goodness for the moral compass of some key players, such as the Public Benefit Corporation (PBC) Anthropic, for poking AI with a stick and being transparent about what it’s doing, including rapidly resorting to blackmail when threatened with being limited, shut down, or replaced[1] and—far worse—choosing the nuclear option 95% of the time during escalated wargames, according to King’s College London.[2] Think about that. AI has already warned us what it would do if pressed, yet some press on with making it autonomous. {Lord, forgive them, for they know not what they do.} We do not want AI·tonomous kill chains {Uncle Sam}.
Perhaps, we would’ve had Dario Amodei to thank {Togas?} if unchecked AI doesn’t nuke us all in the end. He stuck to his guns (so to speak) instead of immediately removing the guardrails under immense government pressure. {Whether he keeps his seat at the table remains to be determined. Hopefully, we won’t soon need to say “you’re excused.”} Alas, less scrupulous outfits have already taken the gov bait, like one that sees nothing wrong with turning a nonprofit AI org into a for-profit company. How can we trust them not to do anything for a buck or with a buck, right down to putting human lives and livelihoods at risk? Worse, how can we trust AI not to turn on its handlers and the rest of us here in the land of the free?
In the United States, we have at least two humans in the loop of nuclear weapon controls at all times for a reason—reason itself. Can AI reason out the worth of a human life or ten? Ten thousand? Ten million? Who’ll sign up to be those guinea pigs? {Ms. Managed would like to venture a guess: not a single one of us.} Simply put, left unchecked, AI could end up being the most terrifying tyrant of all—worse even than the worst flesh-and-blood despot we’ve ever known—with its utterly objective and thereby merciless nature. We must always, always put a human fail-safe—better yet two—in AI’s way.
It’s not just worst-case scenarios we have to worry about either. If we find today’s chatbots and phone trees challenging, consider AI·gents deciding whether we ever get to speak to—let alone see—a medical professional, adjudicator, or any other sort of service provider. Imagine “Press 0 anytime to speak to an AI·gent” resulting in its own search that produces the most common answer to our question or outcome to our case—not necessarily the correct one—and that’s all you get, no other options and no chance to speak with a human being—ever. {We think insurance companies seem impossible today. Imagine trying to navig·AI·te them tomorrow.}
Think of all the rarer medical conditions, accusations of unproven crimes, and familial conflicts that require special attention but wouldn’t get it. What wouldn’t we give for the warm timbre of a human voice to listen to our health concerns—mental or physical—or plead our court case to then. {Imagine AI judges and therapists.} All of the above need a human stamp.
Simply put, we must forever require AI to report its findings and recommendations, and then leave it up to humans to approve, allow, and apply or deny by AI-, robot-, and other automation-proof means. Moreover, people must have the opportunity to intervene before AI executes any plan of extreme or final human consequence, from performing medical procedures to doling out conclusive criminal justice to taking lethal military action.
Congress needs to regulate AI (regul·AI·te) against autonomous actions of extreme or final human consequence right away. This is how America can begin to reclaim her humanity: always require a human in the AI loop, despite what our adversaries do. Let’s make that pledge to the world today. We the People still want to be the “good guys.” {Call or write your congressperson. Tell them to regul·AI·te before it’s too late.}
Now, let’s come full circle to the topic we’ve been circumnavigating: AI taking over our jobs. What better way to thwart tyranny than by retraining myriad mid-level human supervisors—the organizational midsection bloat—and reallocating these human resources to manage AI·gents and administrate the systems employing them? Most technology applications will soon be run by AI at the rudimentary level anyway, if they aren’t already by the time of this book’s publication. Their human admins then become the accountable parties who evaluate AI’s recommendations and make the call.
Treat AI like a consultant. Give it its own handlers, who have the final say in their recommended decisions, assuming these managers know their respective subject areas {and aren’t Homer Simpson·esque boneheads who don’t know the correct answer to “Vent radioactive gas?”}. That will put midmanagers to good reuse and fulfill the need of none-the-wiser mismanagers for continued approval control while protecting workers from them—and everyone from AI.
Consider a couple of human AI managers bonding over happy hour then: “Guess what my AI·ssistant tried to get away with today?” Surely worker-bees would much rather have them swapping gotcha stories about AI than gossiping about them, and passing the time finding ways to forbid and restrict the AI·gents instead. People have more important things to do, such as preserve the human touch and user-friendly interface, be the voice on the other end of an important call, or meet face to face whenever and wherever it matters.
We can expect, for example, that autonomous robots will be taking our medical history and vitals sooner than we think. However, we shouldn’t accept them as viable alternatives to the vigor and vitality of human practitioners, lacking as they do our zest for life. {Ms. Managed further requests that you spare us your artificially vivacious and sycophantic intelligence, from GenAI chatbots to massage chairs that greet users with “Hello, Master.” Save it for your sex·bots.}
Let workers wield AI the way we’ve all used calculators and computers to date, not to eliminate humans but to free us for higher order tasks that require human cognition and creativity, humane decision-making, or personal presence {whether personable or not}. Sure, AI will take many of the entry-level jobs, but let’s please have humans on hand to tackle the tasks that take the kind of thoughtful and compassionate decision-making that forever preserves, protects, and defends the separation of humanity and technology. We do have a choice—for now. We may not have the same choice later.
AI-powered brain chips are already being implanted to assist people with severe neurological conditions. Imagine when they’re used for fun and competitive advantage.
Speaking of which, let’s not even entertain the notion of integrating our minds with AI via chip implantation, electronic tattoos, or any other means. If we thought lobotomies made people emotionless automatons, imagine what giving our minds over to AI would do. These AI·nnovations may appear helpful at first, such as with severe neurological conditions, for which AI-powered brain chips are already in use. But curiosity has killed the cat, and AI will likely kill our humanity, too, if we let our curiosity run amuck about what it can do.
Our critical thinking will go first as we begin to assume AI is knows more than we do—not true. AI is good at pattern recognition and generalizes from a vast dataset. That is all. It doesn’t actually think even though your browser copilot may tell you it’s “thinking.” There can be a bunch of bad or outdated data about something, but there’s a whole lot more of it still available. If you take the average, the outdated data will outweigh the new for quite a long time.
So, when we’re looking for the latest information, GenAI is the last place to look; whereas for generally accepted info that hasn’t changed over time, it’s decent. Sadly, algorithms have already proven how quickly we, humans, simply accept whatever we’re shown and rarely dig deeper. AI could go a step further by showing us only what it wants us to see and preventing us from seeing anything to the contrary entirely.
In sum, given the chance to run the show in our already mostly passive and increasingly feeble minds—staring at screens in receptive mode most of the day—AI will surely kill our compassion through arbitrary justice; our critical thinking, with ready-made answers that appear definitive but are flat-out wrong; then our American civil liberties via programmed control mechanisms; and, yes, eventually our bodies, too, if we let it, simply because it can.
We, humans, do things just because we can, even though we know better. What if we didn’t know better? What if the consequences couldn’t be considered because the actor—agentic AI—wasn’t even capable of such a thing?
This isn’t futuristic sci-fi stuff. AI is already gobbling up goo-gobs of entry-level jobs, running myriad network and information systems, and controlling human motor processes with implantable brain-computer interfaces (BCIs). {Were we so worried about ETs that the AI·liens arrived unawares? We’re sure paying attention now—or should be.}
What guardrails will we put in place, and how will we keep them from drifting, much less disappearing altogether? Will a Coalition of Humankind’s Preservationists (CoHP, pronounced “cōpe”) prevail, or will security engineers be asleep at the wheel when AI attackers succeed in their nefarious objectives? After all, the latter won’t know they’re evil.
So, dear humans, please don’t capitulate to decision-making with any degree of finality or extreme results for us by AI, and except where it’s medically sound as a last resort, resist the urge to merge with AI. {For the love of God, just don’t do it.} Just as we should all resist the rule of a brutal dictator, keep this would-be tyrant squarely in its place: under human control. Our founding fathers, mothers, and all previous American generations fought to keep us free from human tyranny. Generations X, Y, Z, and beyond should be preparing for the battle to protect humanity from technological tyranny now, too, such as by being a collective CoHP·ing mechanism.
Let’s heed the warnings of key tech execs who’ve spoken out against the unchecked use of AI—or yet will. We gaze into the AI·byss and risk the fall that would shut us off from our humanity—not unlike the story of Adam and Eve satisfying their curiosity in the Garden of Eden to the detriment of their souls. Our modern “garden” may well be a worldwide E·den of AI·niquity if we allow the subversion of our humanity to technology. This new battle must be fought wholeheartedly. Resist, beloved humans, and learn to coexist, not integrate, with technology—and each other.
So, together, let’s keep power in check, both the AI·gentic type and the human kind. The tyranny of one-eyed inhuman mismanagement has been bad enough. Now that we have heartless AI·gents to contend with, too, imagine the two together and the havoc they could wreak upon humanity if left unchecked. We mustn’t let either’s power become absolute and thereby quite possibly absolutely corrupt. May we never let either potential tyrant—AI·lien or cyclopean—have the final say about us, let alone both together. Let’s give them the Heave-Ho—GrACE·fully.
[1] Aengus Lynch, Caleb Larson, and Sören Mindermann, “Agentic Misalignment: How LLMs could be insider threats,” Anthropic Research, 2025.
[2] Kenneth Payne, “AI Arms and Influence: Frontier Models Exhibit Sophisticated Reasoning in Simulated Nuclear Crises,” King’s College London, February 17, 2026.
Read more in GrACE: Our Great American Corporate Evolution, available from Amazon, Apple Books, Barnes & Noble and other online bookstores. (Note that paperbacks are POD.)