The AI Shop: Don’t Fight the Vibe, Invoice It (Notes from an Internal Auditor Who Got Tired of…

The AI Shop: Don’t Fight the Vibe, Invoice It (Notes from an Internal Auditor Who Got Tired of Saying No)In which the author, professionally obliged to assess technological risk in the middle of a gold rush, watches three eternally warring engineering tribes unite for the first time in recorded...

The AI Shop: Don’t Fight the Vibe, Invoice It (Notes from an Internal Auditor Who Got Tired of Saying No)In which the author, professionally obliged to assess technological risk in the middle of a gold rush, watches three eternally warring engineering tribes unite for the first time in recorded history, concedes that the business is both well-meaning and armed with working demos, and — instead of building a better gate — opens a till…the till, freshly installed, slightly nervous…Let me open with a disclaimer I mean more than usually: I don’t like writing advice. The internet is drowning in frameworks, playbooks, and men with whiteboards explaining the Seven Pillars of things they’ve done once — and after three essays about robes and incantations, putting on the guru’s hat myself would be a costume change too ironic even for me. So this is not a guide. There is no framework at the end, no acronym, nothing you can print as a poster. These are notes from the workshop: how I’ve been muddling through a problem I suspect a lot of you are muddling through too, written down mostly because the muddling recently produced something that surprised me. Take it as a colleague thinking out loud over coffee — the beverage is back, we’ve reconciled — not as a sermon. The pulpit stays empty.Some context on why this particular problem lands on my desk at all. Part of my job — the part that fits on no business card — is being something like an internal technical auditor. When a technology wants into our house, I’m the one who assesses the risk level, works out to what extent and at what cost it actually solves our collective problems, and translates the answer into words both the engineers and the budget can survive. It’s a peculiar post: too technical for the business to fully trust, too business-adjacent for the engineers to fully claim. On good days I’m a translator. On bad days I’m the man both sides describe, in separate meetings, as “the blocker.”And then came the hypezeit, and the job changed shape entirely. This essay is about how — and about the small, possibly foolish, genuinely surprising thing I’ve started doing about it.I. The Job Nobody Draws on the Org Chart“Trust, but verify.” — Russian proverb, laundered through Reagan, now a job description…the desk where uncertainty gets an invoice…An internal technical auditor — or whatever your organisation calls its equivalent; the role exists everywhere, usually unnamed, usually welded onto some senior engineer’s actual job like a sidecar — does one thing at bottom: prices uncertainty. A new technology, a new vendor, a new architecture arrives, glowing with promise, and someone must ask the unglamorous questions: what breaks, who pays, how do we leave. Not whether the thing is exciting — excitement is abundant and free — but what it costs across its whole life, including the parts of its life that happen at 3 AM and in court.For most of my career, this was a conversation held inside the guild. Engineers proposed technologies to other engineers; the auditor arbitrated between professionals who shared a vocabulary, a set of scars, and an unspoken agreement about what “production-ready” means. The disputes were real — I have refereed holy wars over message queues that ended friendships — but they were disputes among people who all knew, in their bones, that the demo is the beginning of the argument, not the end of it.That world is gone. Not degraded — gone, in the space of about two years. And it’s worth being precise about what killed it, because it wasn’t the technology.II. Then the Business Discovered It Could“The best way to predict the future is to invent it.” — Alan Kay; the business took this personally…an architecture, designed somewhere over Frankfurt…]Here is what the hypezeit actually changed in my corner of the world, and it is not what the keynotes said it would change.AI didn’t just democratise the writing of code. It democratised opinions about the stack. These are profoundly different events, and the second one is the earthquake. For decades, the stack was clergy territory: the business told us what hurt, we told them what we’d build, and the technology choices lived on our side of the rood screen, in Latin. The business couldn’t have an opinion about our message broker for the same reason I can’t have an opinion about transfer pricing — no access to the material.Now? Marketing builds a working prototype over a weekend. A department head returns from a conference with an architecture — not a wish, an architecture, with a diagram, generated in flight, somewhere over Frankfurt. Vendors sell straight to business units, routing around engineering the way rivers route around inconvenient rocks. Every person in the building has, in their pocket, a machine that produces plausible technical artefacts on demand — and possession of artefacts, as any anthropologist will tell you, confers the confidence of expertise, whether or not the expertise came with it.And crucially — write this on the wall, because the entire essay turns on it — they are not wrong to feel empowered, and they are not acting in bad faith. The prototypes genuinely work, in the sense that demos work. The problems they’re attacking are genuinely ours — real friction, real costs, things engineering has had in the backlog since the Obama administration. These are our colleagues trying to help the company we all allegedly care about, using tools that are legal, cheap, and sitting right there. Whatever this essay is, it is not a story about villains. It’s a story about what happens when everyone in it is right.III. The Miracle of the United Guilds“Nothing unites the quarrelling houses like a stranger at the gate.” — some medieval chronicler, probably, watching a demo…the delian league of ops, founding session, catering by panic…And now, the scene I genuinely never expected to witness in my lifetime, and I have witnessed database migrations.To appreciate it, you must understand the ancient geopolitics. Developers, infrastructure, and security are not colleagues; they are three tribes locked in a triangular cold war older than most of our codebases. Developers want to deploy on Friday; infra wants them tried in The Hague for it; security wants both parties preemptively imprisoned and their laptops encased in resin. They agree on nothing. I have watched these tribes fight for a decade over VM sizing, over firewall rules, over whether “it works on my machine” constitutes a deployment strategy. Peace summits have been held over lunch and collapsed by the afternoon standup.And then the business walked in with a weekend prototype and a bright, sincere “look, we can do it ourselves now” — and something happened that no consultant, no reorg, no team-building ropes course had ever achieved.They agreed. Instantly. Unanimously. For the first time in recorded organisational history, the developer, the infra engineer, and the security officer spoke with one voice, and the voice said: “nieee.” Not even “no” — the long, low, involuntary nieee of a parent watching a toddler reach for the hob. It was, I must admit, beautiful. Historians of the corporation will study it. Nothing in the annals of alliance-forming — not NATO, not the Delian League — assembled faster than three engineering guilds confronted with a citizen-built prototype requesting production access.The auditor in me made a note, though, even while enjoying the spectacle. Because I had just written several thousand words, in another essay, about what unanimous reflexive reactions usually are. And this one had all the hallmarks: instant, emotional, arriving before the analysis rather than after it. The guilds had produced a verdict at the speed of fear. Which meant my job, awkwardly, was now to audit both sides.IV. Auditing the “No” (And the Checkmate That Isn’t)“The map is not the territory.” — Alfred Korzybski, actually, this time, for real…two boards, one projector…The guilds’ nieee is not one substance — it’s an alloy, and the metallurgy is quick. Part of it is genuine scar tissue: the migration that ate a weekend, the breach that ate a quarter, the “temporary” system entering its ninth year. That reflex has saved this company money it will never know it didn’t spend — reliability’s eternal PR problem. But part of it, and the previous essays oblige me to say so, is gate defence: if hype is a confession of fear, so is a veto, and three chronically warring tribes agreeing instantly confesses fear for the gate itself. It is Latin defending Latin — and reader, I felt the reflex too; nobody can smell their own incense. Wisdom and territory, alloyed beyond separating, which is precisely why the reflex cannot be the policy. A policy has to tell those two apart. A reflex just fires.And while the guilds were firing, the business made its move — the one you’ve heard too, delivered without malice, which is what makes it formidable: “We don’t want anything bad. We want to solve the company’s problems faster. And everyone can do this now — look, it works.” Intentions clean, tool legal, evidence running right there on the screen. Checkmate?Honour the feeling before dismantling it, because glibness here is how engineering loses in front of the CFO: “everyone can” is true. The cost of producing a working-looking artefact really has collapsed to a weekend and a subscription, and there is no faster way to lose credibility than arguing with a projector. But the checkmate is real only on its board — and the board on the screen is not the board the company plays on. “Everyone can” is true in the dimension of creation and silently false in the dimensions of operation, security, maintenance, and consequences. The king isn’t mated; the king is standing on a different chessboard, in a longer room, where the pieces are on-call rotas and liability clauses and the clock runs for years.Which is why the guilds’ instinct — refute the move — was always doomed. You cannot refute a true statement. You can only complete it: everyone can build. The sentence was never finished, and my entire job, I realised, is the rest of that sentence — which, it turns out, has a price list.But before we get to the till, we need to open the box the business brought. Because what’s inside is instructive.V. Anatomy of the Concept (A Field Report)“Entropy isn’t what it used to be.” — sign in a thermodynamics lab, probably, and also my job…the stack, as rolled, natural twenty on the database…Let me describe, as a composite — details blurred to protect the enthusiastic — what actually arrives on the auditor’s desk when a business-built concept requests residence in production. Not to mock it. To inventory it, because the inventory is the whole diagnosis.No repo…The code lives in a chat history. Sometimes several chat histories, across two accounts, one personal. The canonical version is “the one on Marta’s laptop,” and the concept of canonical version itself lands as fresh news, received with genuine interest.No CI/CD…Deployment is a person. The runbook is that person’s memory. Asked how it gets updated, the answer is a cheerful “I just regenerate it and paste the new version,” which — credit where due — is technically a deployment pipeline, in the sense that a bucket brigade is technically plumbing.Churn built in…The author is an enthusiast, and enthusiasts, God bless them, rotate. The concept has a bus factor of one, and the one is already eyeing a different department. Nobody has asked who inherits, because inheritance implies death, and the demo has only ever been alive.No auditability…Nobody can say precisely what the system does — not from secrecy but from provenance: nobody wrote it, so nobody read it. It was summoned, section by section, and each section worked, and the sum of sections that work is assumed to be a system that works, a theorem for which production offers daily counterexamples.An exotic stack…And here’s the detail I find genuinely fascinating — an exotic stack that nobody chose Ask why this database, this framework, this queue, and the honest answer is: that’s what the model generated. The stack wasn’t selected. It was rolled, like dice, from the latent space’s mood that afternoon. We have technologies in the building right now whose presence traces back to a temperature parameter. Somewhere, a slot machine is making our architecture decisions, and the slot machine doesn’t attend the post-mortems.And threaded through all of it, the finding that matters most — the one that took me longest to see because it hides behind the others: the holes exist because nobody knew how to describe them or what to ask about. Read that carefully, because it’s not the usual complaint. The business didn’t hide the missing secret rotation, the absent access model, the unconsidered data retention. They didn’t skip the questions. The questions don’t exist in their language. “Who rotates the secrets” is not a question a marketing analyst has ever had reason to possess — so it went unasked, so the secrets sit in the source, which sits in a chat log, which sits in an account that will be deactivated when Marta changes jobs. This is not negligence. This is a missing vocabulary. The hypezeit handed everyone the notation and shipped no dictionary of consequences.Once I saw it that way, the entire problem changed shape. You cannot fix a vocabulary gap with a veto. A veto is just another word they don’t have context for.VI. Known Unknowns, Unknown Unknowns, and Unknowable Martas“There are known knowns… there are known unknowns… but there are also unknown unknowns — the ones we don’t know we don’t know.” — Donald Rumsfeld, accidentally producing the finest risk-assessment taxonomy of the century…four quadrants, one flashlight…Rumsfeld was mocked for that quote, which tells you more about the mockers than the taxonomy, because it is precisely correct and I use it weekly.The guilds and the business, it turns out, aren’t separated by intelligence or intent. They’re separated by which quadrant they can see. The business operates in known knowns: the demo works, the problem is real, the tool is cheap — all true, all visible, all on the first chessboard. Engineering’s scar tissue lives in the known unknowns: we don’t know when this breaks, but we know that it breaks, because everything breaks, because we were there the last nine times. That’s the wisdom fraction of the nieee.But the concepts arriving on my desk fail in the third quadrant — the unknown unknowns — and this is the quadrant no reflex reaches. The business can’t see those risks because it lacks the vocabulary (as established); but here’s the uncomfortable symmetry: the guilds’ blanket “no” can’t see them either, because a reflex doesn’t enumerate, it just fires. A veto and a vibe are equally unstructured. Neither one produces a list.And risk management — strip away all the ceremony and the heat maps — is exactly one activity: converting unknown unknowns into known unknowns, one question at a time. Moving items from the quadrant where they ambush you into the quadrant where they merely cost money. That conversion doesn’t happen through gatekeeping and doesn’t happen through enthusiasm. It happens through interrogation — the boring, patient, unglamorous asking of questions the concept has never met.Which is where my actual method finally enters, such as it is.VII. The Audit as a Loaned Vocabulary“I cannot teach anybody anything. I can only make them think.” — Socrates, allegedly, conducting the first architecture review…the dictionary, on loan, late fees apply…]Here’s what I stopped doing: pass/fail. The gate model — concept arrives, auditor inspects, verdict descends — fails on every axis that matters. It kills enthusiasm, which is a genuinely scarce corporate resource, far scarcer than compute. It positions me as the Inquisition, and I’ve written at length about how that costume fits. And, most practically: a verdict transfers no vocabulary. The next concept arrives with the same holes, because “no” contains no information about which questions exist.Here’s what I do instead, and it’s embarrassingly simple: I lend them the dictionary. The audit becomes a structured conversation in which I ask, out loud, with genuine curiosity and zero sarcasm — the zero-sarcasm part is, for me personally, the most expensive line item — the questions their concept has never been asked:Lovely demo — truly. Where does the code live? Who wakes up when it breaks at 3 AM — not “someone,” which person, and do they know? What happens when Marta leaves? Where do the secrets sit, and who rotates them? What data goes in, where does it legally sleep, who can subpoena it? When the model provider changes pricing or deprecates the endpoint, what’s Tuesday like?And then — this is the entire trick, everything else is decoration — I stop talking.Because something reliably remarkable happens. The concept’s owner, who is not stupid, who was never stupid, who merely lacked the words, starts seeing the holes themselves. You can watch it in real time: the demo stops being a finished thing and becomes, before its author’s eyes, a beginning — which is what it always was, but now the author can see the rest of the iceberg, because they’ve been handed the vocabulary in which icebergs exist. Enthusiasm, and this is the part I’d underline twice, mostly survives this. Sometimes it grows — engineers know this feeling well; discovering the true size of a problem is its own dark joy. What dies is not the enthusiasm. What dies is the innocence, and innocence was the risk.The audit stops being a border checkpoint and becomes a language lesson. And once I understood that this was the actual product — vocabulary, not verdicts — the final piece assembled itself, because a vocabulary of consequences has a natural, ancient, wonderfully undramatic representation.A price.VIII. The Three Skeletons in the Closet“In this world nothing can be said to be certain, except death and taxes.” — Benjamin Franklin; production adds pagers…three skeletons, tastefully price-tagged…Every concept that has ever arrived on my desk — every single one, from every department, in every technology, and honestly most of the ones engineering itself produces — carries the same three skeletons in its closet. The details vary; the skeletons never do. So before we open the shop, meet the inventory, because the shop exists to price exactly these three:Skeleton One: Responsibility…Who takes development — not the weekend version, the boring years of it? Who takes maintenance — the patching, the upgrades, the dependency that goes hostile? And the load-bearing question, the one that empties rooms: who answers for it when it blows up? Not “the team.” Not “we’ll figure it out.” A name. Blame, as I’ve written before, is the one thing in our industry that never gets automated — it always, eventually, resolves to a human, and if you don’t choose the human in advance, the explosion chooses for you, and it has terrible taste.Skeleton Two: Money…Who pays — and from whose budget, which is a different question than “is it worth it”? Infrastructure. Operational costs. The security hardening the demo skipped. The audits, internal and — depending on your industry — the expensively external kind. The ordinary servicing, the monitoring, the person-hours of care. The demo cost a weekend and a subscription; that was the deposit, not the price. Somebody signs for the difference, and the concept must know who, before the difference signs for itself.Skeleton Three: The Pager…When it breaks — when, the auditor does not deal in if — whose device makes the noise? Who has supervision, on-call, escalation? And here comes the sentence that changed my relationships with three departments: it can be us. Engineering will gladly hold your pager. According to the rate card.That last sentence is the hinge of everything, because notice what it does: it converts the guilds’ nieee into a number. And numbers, unlike vetoes, can be negotiated, budgeted, compared, and — most importantly — accepted. The moment responsibility, money, and the pager have prices instead of prohibitions, the whole confrontation collapses into something the business has been fluent in all along: procurement.Which brings us, at last, to the shop.IX. The AI Shop“It is not from the benevolence of the butcher, the brewer, or the baker that we expect our dinner, but from their regard to their own interest.” — Adam Smith, opening an internal platform team…aisle three: tokens, pagers, consequences…So here’s what I’ve started building — the thing that recently dawned on me somewhere between a demo and a nieee, and which I offer with all the caveats the epilogue will pile on: stop fighting shadow IT like a possessed exorcist, and open an AI shop instead.The deal, stated at counter height: Use any tool you like — genuinely, any — that passes a risk assessment, which you pay for. Full freedom on the shelf. And at the end of the month, your department receives an itemised bill: your AI usage, your token burn, plus our operational cost for keeping your thing alive, secure, and audited. Welcome. Baskets by the door.That’s it. That’s the mechanism. No gate, no committee, no fifty-page policy nobody reads. A shop: shelves, price tags, a till. And because engineers and budget-holders alike believe receipts more readily than paragraphs, here is what one looks like — a composite, rounded, but proportionally honest:INTERNAL AI SHOP — MONTHLY STATEMENTDept: Marketing • Concept: "LeadWizard" (prototype → supported service)------------------------------------------------------------------LLM API usage (tokens, metered actuals) ................ $23.41------------------------------------------------------------------Risk assessment (one-off, amortised over 12 mo) ........ $180.00Hosting & infra (isolated env, backups, egress) ........ $310.00Secrets management & access control .................... $95.00Monitoring, alerting, log retention .................... $140.00On-call coverage — pager, shared rota, SLA: business-hrs $850.00Security review & patching cadence ..................... $420.00Audit trail & compliance retention ..................... $260.00Incident budget reserve (actuarial, non-refundable) .... $230.00------------------------------------------------------------------TOTAL .................................................. $2,508.41of which "the AI part" ................................. 0.9%------------------------------------------------------------------The demo was free. The system never is.I keep a laminated copy. Notice what the receipt argues without my saying a word: the tokens — the part everyone debates, benchmarks, and writes think-pieces about — are a rounding error, under one percent of the true monthly cost of the thing existing responsibly. Everything above the line is the demo; everything below it is the system. Twenty dollars of intelligence wrapped in twenty-five hundred dollars of adulthood. No slide deck I have ever produced has communicated this in under five minutes; the receipt does it in five seconds, and nobody has yet argued with it, because it is not an opinion. It is arithmetic with a staple in the corner.And the reason I’ve fallen for the mechanism — provisionally, experimentally, see epilogue — is that it does three things no gate has ever done for me:First, the enthusiasm survives intact…Nobody hears “no.” Nobody is sent to the Inquisition. They hear “yes — here’s the price list,” which is the most respectful sentence one adult can say to another, because it treats them as someone capable of making their own cost-benefit decision. The business wanted democratisation? This is democratisation — the full, grown-up version. Democracy, it turns out, comes bundled with a tax system; every adult in the building already knows this from April.Second — and this is my favourite property — the skeletons walk out of the closet on their own…I no longer have to argue that maintenance costs money; maintenance has a shelf price. I don’t have to prove the pager is a burden; the pager has a monthly rate. Roughly half the concepts — early data, tiny sample, do not cite me — quietly retire at the moment of pricing, not because anyone vetoed them but because their own sponsor looked at the invoice and did the arithmetic that the demo’s glow had been outshining. And the ones that survive the price list? Those turn out to be the genuinely valuable ones — because a price list is a better filter than a committee. You can argue with a committee. You can charm a committee. Nobody has ever charmed an invoice.Third — for my economist readers, both of you — this is just internalising externalities……and yes, Coase is back, fourth essay running; at this point he’s practically a co-author and should share the therapist. Shadow IT flourishes for one clean economic reason: its benefits are visible and concentrated (the department gets its tool, today) while its costs are invisible and diffuse (risk, maintenance, and security land on engineering, later, unpriced). Of course it grows — you’ve built a subsidy for it. The shop simply moves the cost to where the decision is made. Same freedom, same tools, same enthusiasm — but now the person choosing the risk is the person paying for it, which is the only configuration in which the word “choice” means anything at all.And a quiet bonus, fourth thing, almost shy: the guilds get to stop being the villains…Infra and security transform overnight from the Department of No into vendors — people with services, rate cards, and satisfied internal customers. I have watched a security engineer explain his on-call pricing to a marketing lead with the cheerful pride of a craftsman, and I confess it moved me. Nothing rehabilitates a priesthood like giving it an honest shop to keep.X. The Counter-Argument I Can’t Fully Dismiss“For every complex problem there is an answer that is clear, simple, and wrong.” — H.L. Mencken, reviewing my shop from across the street…the author, auditing his own till, finding irregularities…Tradition demands I now rob my own till, and this essay’s counter-arguments have unusually sharp teeth, because the shop is young and I can already feel where it might bite me.The price list can become a veto in a wig…This is the failure mode that worries me most, because it’s the one I’d be tempted by. Price the risk assessment punitively, pad the operational rates, and congratulations: you’ve rebuilt the gate, except now it’s dishonest — an Inquisition with a cash register, prohibition cosplaying as economics. The guilds, who staff the shop, have every territorial incentive to drift this way, one “adjusted” rate at a time. The only defence I know is transparency ugly enough to hurt: published rates, benchmarked against external equivalents, challengeable. The moment the shop’s prices can’t survive comparison with the open market, the shop has become a racket, and the business — correctly — will route around a racket. Which leads to —The credit card exodus…The shop only works if it’s cheaper and easier than the alternative, and the alternative is a corporate card and a vendor who asks no questions. I can price honestly and still lose to a SaaS checkout page that takes ninety seconds. The counter is friction asymmetry — make the compliant path genuinely the lazy path — but I won’t pretend that battle is won by economics alone; it’s won by the shop being actually good, which is a much taller order than being merely right.The agency ambush…And the exodus has a far more expensive cousin, one every internal engineer will recognise with a small facial tic: the business, routed around or priced out, commissions the prototype from an external agency. This is the worst timeline, because it combines every pathology in this essay and adds an invoice with a logo. Agencies are structurally optimised to deliver and depart — they bill handsomely for the build, hit the demo milestone, collect, and exit stage left, and the moment the champagne is cleared, the artefact lands where such artefacts always land: on internal engineering, which now owns the maintenance, the security holes, and the technical debt of a codebase it never chose, in a stack the agency’s own slot machine rolled, with the original authors now billing a different client entirely. It’s shadow IT with a mark-up and a handover document written in the past tense. The shop, if it works, is the shield here — and in two ways: the honest internal rate card gives the business a real comparison before the contract is signed (agencies look cheap only when the internal alternative has no visible price), and shop policy closes the loophole from the other end: external builds don’t skip the till — the same risk assessment, the same operational rate card apply, so “but the agency built it” exempts nobody from the cost of it existing. Whether this holds against a sufficiently determined VP with a sufficiently persuasive agency salesperson, I’ll report back. I am not serene about it.Bureaucratic calcification…Today the risk assessment is a conversation and a checklist. Give it three years of unattended corporate physics and it will metastasise into a forty-field form with a SharePoint workflow, and we’ll have recreated the thing we fled, with extra steps. Processes, like code, rot toward ceremony unless someone aggressively deletes. I’ve appointed myself the deleter. Ask me in three years whether I survived.And the deep one, the one I chew at night: the shop prices the risks we know how to name. But this whole essay established that the defining hazard of the hypezeit is the unknown unknowns — and you cannot put a price tag on an item you can’t see. The shop converts named risk into cost beautifully; it does nothing, by itself, for the unnamed. My hope — and it is a hope, not a claim — is that the vocabulary loop closes over time: every incident, every audit, every retired concept adds words to the dictionary, and the dictionary reprices the shelf. But I’m describing a flywheel I’ve only just started pushing, and flywheels have a documented history of being heavier than their inventors’ blog posts suggest.Confidence rating, per series convention: I’m at maybe sixty percent that the shop is the right shape, and thirty that I’ve got the prices right — measured, as always, after coffee, which the reader should mentally deflate by ten.XI. The Sinusoid, Retail Edition“The thing that hath been, it is that which shall be.” — Ecclesiastes 1:9, still undefeated, now with loyalty points…the spreadsheet that runs the company, 1997–present, uninterrupted…Regular readers know the drill: no essay leaves this workshop without its sinusoid, and this one is a proper vintage, because shadow IT is not new. Shadow IT is one of the oldest waves we have.Excel was the vibe coding of the 1990s. I mean this with complete precision, not as a quip: a tool that let the business build working software artefacts without engineering, without repos, without review — and build them they did, by the million. Entire companies ran, and quietly still run, on spreadsheets of terrifying sophistication, maintained by one person named in no org chart, deployed by email attachment. Then Access, and the departmental database with forms. Then VBA macros metastasising through the finance department like a beautiful mould. Then no-code, low-code, Airtable empires, Zapier plumbing. Every single generation of it followed the same arc: business discovers it can → guilds cry nieee → prohibition fails, comprehensively and always, because prohibition against a cheaper tool has never once worked in the history of tools → the thing proliferates in the dark, holes and all → something explodes expensively → and then, in the rubble, IT finally builds the thing it should have built on day one: a managed path. Governed spreadsheets. Sanctioned low-code platforms with SSO and backups. A shop, in other words — every cycle ends with a shop, and every cycle begins with somebody refusing to open one.Which reframes my little AI shop, with suitable humility: it is not an invention. It’s an attempt to skip ahead in a well-documented loop — to open the store during the gold rush instead of after the mine collapse, on the wild theory that the pattern being visible might, for once, exempt us from completing it. The sinusoid’s entire track record says patterns are never skipped, merely accelerated. But acceleration would already be worth the till.XII. Epilogue: A Loose Thread, Deliberately Left Loose“I have not failed. I’ve just found 10,000 ways that won’t work.” — Edison, again, because this essay is honest about which stage it’s at…open sign, hand-lettered, slightly crooked, sincere…And now the ending this essay owes you, which is not a conclusion, because there isn’t one yet.Everything above — the shop, the price list, the skeletons with tags — is a loose concept. I’m not reporting a proven system; I’m sketching something that recently, and rather suddenly, dawned on me — one of those moments when you’ve been pushing on a door for a year and someone points out the hinge. It felt like illumination. I am old enough to know that feelings of illumination have a base rate, and the base rate is unkind. There are surely other approaches — better ones, probably, being run right now by quieter people who don’t write five thousand words about it; if you’re one of them, the comments exist and I read them, as this entire series can attest, occasionally at the cost of a month.What I can tell you is only this: I’m trying it. The till is open, the rate card is drafted, the first concepts are being priced, and the first sponsor has already done the silent arithmetic and withdrawn a project with, I want to note, his enthusiasm and his dignity fully intact — which is one more graceful exit than the gate model produced in five years. Whether the shop survives contact with corporate physics, whether the prices stay honest, whether I’ve merely built a prettier gate — I genuinely don’t know. Ask me in a year. There may be a follow-up essay; there may be a retraction; knowing this series, there will be both, and a sinusoid.But if there’s one thing I’d keep even if everything else burns down, it’s the reframe underneath: the hypezeit cannot be refused, only priced. The business is right that it can. The guilds are right that it costs. The auditor’s job — my job — turned out not to be standing between those truths, but introducing them to each other, itemised, at the counter.The shop is open. Mind the skeletons. They’re by the till, and yes — they’re for sale too. Everything here is. That was always the point.Krzyś, aka The Author, aka the man behind the till, remains a full-stack engineer, part-time Stoic, and now — apparently — a shopkeeper, a career development no aptitude test ever predicted. He continues to audit concepts by day and his own counter-arguments by night, and reports that the guilds now bring him coffee, which after a decade of being called “the blocker” feels frankly disorienting. His therapist, upon hearing that he has started charging colleagues for risk, paused for a long time and then asked for the rate card. He considers this his most successful knowledge transfer to date.This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories.Subscribe to our newsletter and YouTube channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!The AI Shop: Don’t Fight the Vibe, Invoice It (Notes from an Internal Auditor Who Got Tired of… was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.

Source: Generative AI Pub — Published July 13, 2026 at 10:52 UTC — Category: Image AI