Oh, look pappy! Another buzzword-ridden article on AI governance.
Yes-sure-ee son!
Yeah, no thank you.
But I did feel the urge to challenge myself to write about something so completely alien to me, and still be able to keep my eyes open while doing it.
It’s a challenge.
Do you know what actually a “challenge” feels like?
Yeah well, it ain’t endless discussions on corporate policies, compliance frameworks, and validation hierarchies, that’s for sure.
Cause, let’s face it, governance is the fiber-rich, vegan-approved, got-milk promoting cornflakes of the AI world. It is essential for avoiding disastrous organizational constipation, but about as thrilling as a hot-’n-damp Tuesday afternoon meeting in the Nether-Regions where I live, on how to keep our feet dry coming summer.
Still here?
Fine.
Let’s trudge through this together.
We can do it, you and I!
Here we go. . .
Implementing effective governance for AI applications is crucial for enterprises that wish to harness the power of artificial intelligence responsibly.
You sure about this?
Nah, me neither.
I had ChatGPT write this for me. And reading this alone proactively gave me yawning cramps.
So let’s do it different.
More rants after the commercial brake:
- Comment, or share the article; that will really help spread the word 🙌
- Connect with me on Linkedin 🙏
- Subscribe to TechTonic Shifts to get your daily dose of tech 📰
- Visit TechTonic Shifts blog, full of slop, I know you will like !
In the beningging
Anyways. . .
In the beginning, the first form of governance was born out of true necessity.
I think it had to be a school somewhere (in Alabama of course), where teachers were always trying to stop toddlers from shoving crayons up their noses. So they armed themselves with policies, procedures, and desperately hoped the whole thing wouldn’t blow up spectacularly, straight into their pesky little brains.
At first, the governance was straightforward. “No crayons in noses”, the teachers wrote firmly on brightly colored posters decorated with smiling cartoon characters.
But toddlers being, well, toddlers, much like AI algorithms, are inherently creative, wildly unpredictable, and determinedly resourceful. So, soon enough, the crayons weren’t the problem anymore, because now it was beads, marbles, erasers, and the occasional finger paint-filled glue stick.
Panic set in.
But teachers being teachers, they reacted swiftly, and they created a new hierarchy of oversight. On the one hand they appointed classroom monitors to watch potential offenders, assistant teachers to monitor the monitors, and a stern-faced vice principal who oversaw the entire surveillance operation from behind a windowless office door, and eventually, the crayon-free policy grew into a 300-page document, complete with detailed diagrams and exhaustive appendices specifying the permissible distance between toddlers and coloring implements.
But despite this impeccably well detailed, and thoroughly executed governance framework, things still went spectacularly sideways.
Again, because toddlers, like rogue AIs, found loopholes.
They discovered crayons hidden in forgotten backpacks, and they started to trade contraband markers behind the jungle gym, and even staged raids on the art supply closet. The school responded with further restrictions, random backpack inspections, frisking in the schoolyard, and instilling mandatory safety drills, but the toddlers always stayed one step ahead.
And in the end, the teachers became exhausted and disillusioned, and then they realized something important. . . perfect governance was impossible.
The chaos is just too inherent, and creativity too resilient, and human nature (especially toddler and the occasional Gremlin) is too resistant to absolute control, so they scaled back the rules, and suggested a compromise, and accepted that sometimes, despite their best intentions, crayons might still end up lodged where they shouldn’t.
And perhaps, that’s the real, cynical truth at the heart of all governance, whether dealing with toddlers, corporations, or AI – absolute control is an illusion.
Governance is merely our hopeful attempt to minimize disaster, when we know full well that we’re only one determined toddler, or one misfiring AI, away from a colorful disaster.
The end.
The real beginning
Chucks, I could’ve just stopped here, and perhaps I should?
Well, me being me, I know I won’t. So if you really want to learn some practical skills – don’t stick around, and go implement your own governance – but if you want to have a good time, and learn something about governance while at it, well, stick around my (remaining) friend.
In our grown-up case, there’s little less crayons and potential brain damage (for the most of us), but the gravity of the situation is as dire as it was in those schools. With us Grups, working with AI, Governance means we need to lay down the law about how our organization’s AI uses, manages, and generally abuses data.
You see, old-school data governance was neat and tidy, all structured schemas and predictable outcomes, kind of like your grandmother’s knitting circle, only with less gossip and more SQL databases. BUT AI governance, by contrast, is like trying to manage a circus full of cocaine-addicted chimps, cause there’s no schema, no certainty, and just algorithms making wild (probabilistic) guesses based on mountains of unstructured junk (data), and all the while we are hoping that whatever emerges isn’t totally insane.
Still, here you are (said Yoda), and you’re still naively believing we can keep these algorithmic beasts in check?
Adorable, but let’s continue.
Real “practical” (by which I mean slightly evil, or practevil) AI governance is simple in principle but an absolute migraine-inducing nightmare in reality.
So, let’s get this straight before we continue…
Memorize kids: “All governance is inherently evil”.
Répète (French):
- “All governance is inherently evil”
- “All governance is inherently evil”
- “All governance is inherently evil”
Thank you.
Just to check if you memorized it. . . what is governance about?
Evil, yes !
Hmm. . . maybe that’s not supposed to be the message of this post.
But I digress.
Well, in practice it luckily doesn’t involve creating an army of faceless bureaucrats, digital or otherwise, whose sole purpose in life is to say “no” to interesting or cool things and “yes” to painfully dull procedural . . . documentation, but what it does mean, is erecting countless validation layers, approval hierarchies, and corporate rituals resembling ancient blood sacrifices, which are all aimed at preventing your AI from making your entire company implode spectacularly on a random Tuesday.
And I am not going to talk any further about this particular form of governance – the governance that hoomans do onto one another, even if it has a fancy sticker called AI slapped onto it.
I said it.
So, people with a short attention span, you can go take your pills now, or go to bed.
As I see it, real-practical governance in the AI age works on a philosophy of layered despair.
One layer is the hooman layer, which I am not going to elaborate about. . go Google some 40 page document on the topic, try getting your head around one of them bloated EU AI-governance frameworks, or have your trusty ChatGPT whisper it in your ear while you sleep, see if I care.
No, I will not be talking about that layer.
But what I will be talking about is the second layer, the AI-layer or Agentic AI-layer itself.
Agentic AI-GovernanzZ
Now here is where it really starts getting practical!
You see, every AI model is, in its heart, a non-deterministic chaos engine, much like Marc Drees in his writings.
Now that’s a whole mouthfull to digest, and I get it if you wish to quit, and I encourage you to quit even, if the term non-deterministic doesn’t ring a bell, because in that case you should absolutely not have a say in determining what your organization’s AI-Governance is supposed to look like anyway.
Say, you trained it on tons of data, but half of it came from Reddit posts, badly-written company emails, and expired PDFs nobody’s opened since 1998.
By now we all know that each AI-generated response is essentially a hopeful guess masquerading as profound wisdom brought to you with pure hubris, and to manage this beautiful chaos, companies deploy what they call “domain-specific agents“, now that’s a fancy term for little digital bureaucratic shits who specialize in certain tasks, and who, at least in theory, prevent the whole circus from collapsing into complete anarchy.
Sounds exhausting, and that’s the point.
It’s supposed to be exhausting.
Aye, I be feelin’ it in me bones as well, matey! (said the pirate).
It is supposed to drain you, dear Jack (Sparrow).
Welcome to the real world of AI governance, on an agentic scale.
Now allow me to take you deeper into the bureaucratic depths. You see, Agentic AI Governance thrives precisely because it’s fueled by mistrust, suspicion, and a seemingly infinite capacity for creating. . . even more bureaucracy.
Next, let me introduce you to your new favorite dysfunctional member of the agentic AI-governance family, the Executors and the Validators. Executors are your optimistic, fresh-faced agents who are tasked with providing you brilliant answers and practical solutions. They generate documents, recommendations, make a few reports, and go to work with tools, and all of them are naive enthusiasts and are blissfully unaware that they’re about to be torn to shreds. Basically, Executors are the ambitious, hopeful, hard-working, and tragically disposable members of the family.
And then we meet their sinister cousins, the Validators.
No, not the Terminators. I know you !
These agents are all about skepticism, pessimism, and sheer bureaucratic evil. They’re the permanent critics, suspicious and reviewing each output of a hopeful executor as if it were evidence in a murder trial. Validators are those colleagues who always start sentences with “Well, actually…”, or “Have you included all of our stakeholders?” and whose greatest masochistic joy in life is finding tiny mistakes that utterly demolish the executor’s credibility. Validators exist solely to mock executors’ optimism by flagging questionable phrases, and mercilessly strike through any responses deemed insufficiently dull or suspiciously insightful.
Validators are supported by another, even more tedious set of agents, the Auditors. Oh my good lord, yes, even AI has auditors, and yes, they’re exactly as thrilling as their human counterparts.
The auditors look at every single interaction between executors and validators, and look for the tiniest bit of noncompliance to arbitrary policies written years ago by people who no longer even work at your company.
And auditors live to nitpick.
Their very (digital) existence revolves around maintaining this endless chain of suspicion. They create mountains of digital paperwork that no human ever willingly reads, but which must, nonetheless, be dutifully stamped “approved” before being filed away and forgotten.
If this sounds exhausting, good.
Remember, I warned you.
But lemme tell ya about the secret of Agentic AI Governance. . . it actually works, well, sort of.
The battle between executors, validators, and auditors creates a perpetual state of tension that (in theory) drives the quality of AI-generated content upward, making the outcomes less dangerously ridiculous. This carefully maintained tension, this artificial friction, is precisely what keeps your organization’s AI from confidently asserting things like “Trump farts fire”, or worse yet, “the client will definitely sign that deal, not!”.
But, hold your applause, because there’s more!
Jeez.
Lots more.
Just have a look at the sheer amount of text we already covered. And yes, talking about agentic AI is as tiresome as the hooman counterpart, but still we have to, because AI governance wouldn’t be complete without its greatest, most unnecessarily elaborate trick, and that’s the Hierarchy of Validators.
Dawat?
Sounds like a good name for a Warhammer movie.
The thing is, that we should not just trust a single validator, we stack them in elaborate, dizzying hierarchies, where each layer is more suspicious, and each validator more cynical than the last.
Validators validate other validators, in endless loops of self-doubt and digital existential crisis, all trying to prove their ruthless efficiency by tearing apart the work of their fellow agents.
This is what Russia must have been like under Stalin and Beria.
But luckily for us, this takes place at the quantum level.
And finally, at the heart of it all, there is knowledge governance.
It’s a short one, so stay alert.
Knowledge GovernanzzzzZ
You see, AI agents rely on lotsa data to function, but the truth is, that your company’s knowledge databases are overflowing with junk. Ever heard your, say, marketing department complain about the quality of data?
Yuuup.
That’s what I’m talking ‘bout.
Outdated documentation, contradictory guidelines, and PowerPoint presentations that no sane person has ever read fully. Agentic AI Governance solves this not by actually tidying up this mess, but by creating another layer of agents whose entire existence is to certify these dusty files as “official knowledge”. Think of these bots as being librarians who are sorting books that no one will ever check out again.
It is so magnificently futile, yet beautiful in its sheer pointless extravagance.
But despite the absurdity, despite the evident futility, we carry on.
We go on because governance is, ultimately, about plausible deniability.
Here, I said it.
PLAUSIBLE DENIABILITY.
That basically sums up what governance is all about.
Our corporations create layers and rules and validations not because we genuinely believe we can completely control AI’s chaotic potential, but because we want to pretend we tried everything humanly, and digitally, possible when things inevitably go wrong.
In that way, governance isn’t about preventing disaster, but more about assigning blame effectively when disaster inevitably strikes.
Let that sink in.
And that’s why good C-level executives do NOT hide behind governance policies to make a decision, they use their knowledge, their experience and their intuition to make a judgement call and let the minions mop it up if it fails, because they know that in the end, no policy, nor procedure can take the blame. Only a hooman can. And luckily I have worked with a few of those people who dare to make executive calls in large corporations, without hiding their peachy asses behind paper walls.
So, here’s your practical guide to Agentic AI Governance in a nutsack (hahaha, I know for sure no one got this far anyway, but if you happen to wake up and stare at the last paragraph, and stumble on this lil gem – kudos! – let me know in the comments, by repeating the word backwards).
Signing off thousands of documents, and finally hitting the (not)sack.
Marco
I build AI by day and torch it by night. I call it job security. Let’s keep smashing delusions with truth. We are the chaos. We are the firewall. We are Big Tech’s PR nightmare.
Think a friend would enjoy this too? Share the newsletter and let them join the conversation. Google and LinkedIn appreciates your likes by making my articles available to more readers.
To keep you doomscrolling 👇
- The AI kill switch. A PR stunt or a real solution? | LinkedIn
- ‘Doomsday clock’: it is 89 seconds to midnight | LinkedIn
- AIs dirty little secret. The human cost of ‘automated’ systems | LinkedIn
- Open-Source AI. How ‘open’ became a four-letter word | LinkedIn
- One project Stargate please. That’ll be $500 Billion, sir. Would you like a bag with that? | LinkedIn
- The Paris AI Action summit. 500 billion just for “ethical AI” | LinkedIn
- People are building Tarpits to trap and trick AI scrapers | LinkedIn
- The first written warning about AI doom dates back to 1863 | LinkedIn
- How I quit chasing every AI trend (and finally got my sh** together) | LinkedIn
- The dark visitors lurking in your digital shadows | LinkedIn
- Understanding AI hallucinations | LinkedIn
- Sam’s glow-in-the-dark ambition | LinkedIn
- The $95 million apology for Siri’s secret recordings | LinkedIn
- Prediction: OpenAI will go public, and here comes the greedy shitshow | LinkedIn
- Devin the first “AI software engineer” is useless. | LinkedIn
- Self-replicating AI signals a dangerous new era | LinkedIn
- Bill says: only three jobs will survive | LinkedIn
- The AI forged in darkness | LinkedIn
Leave a Reply