A true story about office furniture, professional fear, and a prompt that changed a Tuesday morning
Author's note: The characters are composite fictions. The town hall, the school, and the county are real as types, not as identities. Any resemblance to living civil servants is, in equal measure, unavoidable and deliberate.
The Tuesday with Furniture
At 8:17 on a Tuesday morning, Gheorghe Ciubotariu opens the window of his office at the town hall in Hliboca — not the Ukrainian Hliboca, but the Romanian one: a market town of a few thousand souls separated from the border by twelve kilometres of flat farmland and a time zone that nobody chose. Through the window comes the smell of wet linden blossom and the sound of a tractor. On his desk: a mug of instant coffee, a blue Schneider biro with a chewed cap, and the procurement requisition that the mayor flung on his table on Friday with the words: "Gheorghe, sort it. By Thursday."
The requisition has three lines. Office furniture — chairs, desks, cupboards, shelving units. The council chamber and the secretariat. Estimated value somewhere below the equivalent of twenty-six thousand euros — which means a direct purchase or, if anyone has been adding items, a simplified tendering procedure. The mayor is a newcomer from the private sector who has suddenly noticed that the council chamber looks like a 1994 dental surgery and that, on the 15th of July, a county delegation is coming to visit.
Gheorghe is fifty-four years old. He has been a public procurement adviser for nineteen years — that is to say, before SEAP (Romania's national e-procurement platform), before e-licitatie.ro, before anyone had heard of Law 98/2016 on public procurement. He has survived three iterations of procurement legislation, two reorganisations of ANAP (the National Agency for Public Procurement), the transition from paper files to scanned files to uploaded files, and every platform update that arrived with usage instructions dispatched on Friday evenings at 17:58.
Gheorghe does not hate technology. He respects it the way a mountain person respects an unfamiliar path: with attention, with scepticism, and with the awareness that one wrong step is enough.
The Paper That Protects
He knows what he has to do. He knows the procedure by heart — not as a recited regulation but as a lived ritual. The technical specification. The product data sheets. The CPV codes. The justification note for the estimated value. The requisition signed by the mayor. Invitations to tender sent to a minimum of three suppliers. The response deadline. The evaluation committee. The minutes of proceedings.
He opens a blank folder and writes on the cover, in blue biro: "Office furniture — July 2025".
The first thing he does is not to write. It is to look. He opens the archive — a beige-painted metal cabinet with a key — and retrieves the file from the last similar procurement, in 2022. The technical specification from that year. The justification notes. He lays them on his desk. He compares the prices from then with what he knows has become more expensive. He adds a percentage in his head. He adjusts.
He is a methodical man. He is not idle, not incompetent. He is conservative in the most honest sense of the word: he has seen enough files sent back by ANAP, enough challenges won by irate suppliers, enough errors that cost someone their post or their freedom. He knows that incorrect documentation does not mean a fine — it means a criminal investigation. And he knows that simplification is a trap: whatever you cut short today, you repay tomorrow with interest.
At 10:30, the mayor enters without knocking. He sees Gheorghe with the 2022 file open and a blank sheet in front of him on which roughly six lines have been written.
— Gheorghe, by Thursday, you said? Tuesday is Thursday.
— Well, Mayor, the documentation...
— What documentation? It's furniture, not nuclear reactors.
Gheorghe wants to explain that this is not a matter of complexity but of procedural correctness. That if you enter the wrong CPV codes, the catalogue does not return the right suppliers. That if the estimated value is calculated poorly, you risk being below the threshold when you are actually above it. That response deadlines also have legal logic behind them.
But the mayor has already left.
The Phone Call
At 11:15, Gheorghe calls the local primary school. He is looking for Mihai Cojocaru — the school's accountant, a man he has known for twenty years, from football and from weddings. Mihai is about eight years younger and has a reputation as someone who gets things done.
— Mihai, did you do anything with furniture in the spring?
— I did, says Mihai. Kitting out a reception class. Desks, chairs, a cupboard. All below the direct-purchase threshold.
— And how long did the documentation take you?
A pause.
— About an hour.
Gheorghe thinks he has misheard.
— An hour?
— An hour, maybe a bit more. I fed everything into an AI platform. I described what I needed, what my budget was, which procedure applied, it asked me a few things, and it gave me the technical specification, the justification note, the product data sheets with CPV codes, a draft invitation to tender — the lot. Then I checked it myself, adjusted two codes, and published.
— And it was legally sound?
— Legal, transparent, zero challenges, contract signed in eighteen days.
Gheorghe goes quiet. He listens to the tractor outside. He looks at the 2022 file.
— Send me what you used, Mihai.
— I'll send you the whole file and the link. But Gheorghe — feed them into the platform. Don't copy them. Use them as context and tell it what you want, for your specific situation. Otherwise it won't work properly.
The Link
At 11:47, Gheorghe receives an email from Mihai with the subject line: "how I did it". The contents: the school's scanned procurement file — eight PDF documents — and a link. At the bottom of the message, Mihai has added a postscript: "I went to a knowledge-sharing event in Cluj last month. Someone from a Bucharest district council and someone from a digitalisation NGO showed us how they use these platforms. The trip was worth it."
Gheorghe opens the link. The platform is clean, white, with a text box in the centre. The cursor blinks.
He stares at the text box for approximately three minutes.
What Mihai Saw in Cluj — The Knowledge Exchange
[Narrative insertion — Mihai's voice, reconstructed from what he told Gheorghe afterwards, over a beer]
The room was small, on an upper floor of an old building in central Cluj. Around twenty people — town hall secretaries, public institution accountants, an administrative officer from a county council in Transylvania. Mihai was the only one from the north of Moldavia.
The man from the Bucharest district council — a fellow of about forty, jacket but no tie — put a table on the screen with three columns: What we do today. What process automation means. What implemented AI means.
— Ladies and gentlemen, he said, before we begin I want to clear up a confusion that is costing public money across Romania. People assume that automation and AI are the same thing. They are not. And if you don't understand the difference, you end up buying a locomotive when you needed an aeroplane.
He pointed to the first column.
What we do today — and he described exactly what Gheorghe does: open the 2022 file, copy the structure, fill in the new fields, check it by hand, print, sign, upload. "This is not automation, and it is not AI. This is manual work with digital tools. It's the equivalent of writing with a pen, but on a tablet."
Process automation — RPA, he said. Robotic Process Automation. A software robot that does exactly what the person does, but without the person. Opens the application, fills in the fields, saves, sends. Fixed rules, predefined steps, no judgement. "Estonia started with this twenty-odd years ago. X-Road — their national data-exchange infrastructure — is essentially process automation at a state scale: a citizen submits a request once, and the system routes it automatically to every institution that needs it. No one carries a piece of paper from counter to counter."
Mihai wrote in his notebook: Estonia — a citizen submits their data once. In Romania — how many times?
Implemented AI — and here the man from Bucharest slowed down, because the room had grown perceptibly more attentive. "Implemented AI does not mean you have bought a ChatGPT licence and use it to write emails. Implemented AI means the system judges, not just executes. It means the platform reads a procurement requisition, understands the context, selects the correct procedure from among several possible options, generates documents tailored to the specific situation, checks them against an up-to-date legal database, and asks clarifying questions when something is ambiguous. It means the system can do what Gheorghe does — not what a robot does."
Someone in the room asked: But doesn't that mean Gheorghe is no longer needed?
— No. It means Gheorghe's work changes. If Gheorghe checks and judges — he stays. If Gheorghe merely copies and fills in — he disappears. But it is not the platform that replaces him. It is someone else who knows how to use the platform.
Three Countries, Three Lessons — What This Looks Like Elsewhere
[The narrator's voice. Mihai brought home three pages of notes. He photographed them and sent them to Gheorghe on WhatsApp.]
Estonia: infrastructure as the state itself
Estonia did not "implement AI". It rebuilt the state as digital infrastructure, and then added AI on top. Its X-Road platform now connects more than nine hundred public institutions in a data-exchange system where a request submitted once arrives automatically at every relevant entity — no phone calls, no visits, no files carried by hand. On top of this infrastructure, the Bürokratt system functions as a state virtual assistant: citizens ask questions in natural language and receive answers that navigate real legislation and real procedures. The plan for 2026 includes personalised AI agents and an Estonian-language LLM built by the state itself, not purchased from an American corporation. The name of the AI strategy — Kratt — refers to a mythological creature in Estonian folklore that performed menial tasks for its master. The bureaucratic poetry is not accidental.
The lesson: They automated data flows first. Then they put AI on top of them. Not the other way around.
Singapore: "It is no longer optional"
In Singapore, GovTech — the government's technology agency — announced in 2026 something that in Romania would sound like science fiction: they intend to provide one agentic AI assistant to every public officer, to support daily tasks. Not a chatbot that answers questions. An agent that can plan, execute actions across systems, monitor progress, and report. The agency's chief executive stated it plainly: "Not using AI is no longer an option."
They have traversed three distinct stages, which they have also documented: first, AI tools for individual productivity (writing documents, analysing data); then centralised government-wide AI procurement platforms; now, agentic integration — systems that do not merely assist the person but work alongside them across complete end-to-end processes.
The lesson: Adoption happened in stages, not in a single leap. Each stage prepared people for the next.
The United Kingdom and the United States: large budgets, substantial fragmentation
The United Kingdom spent £573 million on government AI contracts by August 2025 — more than the whole of 2024 combined. The United States committed $5.6 billion to AI between 2022 and 2024. And yet independent assessments show that US states perform worst of all on building internal capacity: they buy AI, but do not know how to evaluate it, adapt it, or audit it. They purchase locomotives and leave them in the depot.
The lesson: Money solves the budget problem. It does not solve the mindset problem. These are different problems.
Back to the White Box
Gheorghe types the words "Good morning" — and deletes them. He types again:
"I am a public procurement adviser at a town hall. I need to prepare documentation for the purchase of office furniture — chairs, desks, cupboards, shelving units — for the council chamber and the secretariat. The estimated value is below the direct-purchase threshold, so the procedure is either a direct acquisition or, if the value is slightly above, a simplified tendering procedure. I need: a technical specification with product requirements, a justification note for the estimated value, a requisition form, the correct CPV codes, and a draft invitation to tender to be sent to three suppliers."
He presses Enter.
What follows does not resemble anything Gheorghe has previously seen on a computer screen. It is not a form. It is not a drop-down menu. It is a conversation — but one in which the interlocutor has read all the legislation Gheorghe knows by heart, and has also understood it in context.
The platform asks three questions: the number of workstations in the secretariat (to size the furniture correctly), whether there is a material preference (MDF, melamine-faced chipboard, metal), and whether there are colour requirements or compatibility constraints with an existing inventory register.
Gheorghe answers in two sentences.
By 12:34, he has on screen a complete technical specification with product requirements and CPV codes — 39130000-2 (office furniture), 39110000-6 (seating), 39132100-7 (filing cabinets) — a justification note citing the relevant provision of Government Decision 395/2016, a standard-format requisition with completed fields, and a draft invitation to tender with the legally required minimum response period.
Gheorghe reads everything. He checks the CPV codes against the SEAP catalogue — correct. He checks the legal reference in the justification note — accurate. He checks the deadlines — compliant.
By 13:10 he has sent invitations to three suppliers, published the announcement on the town hall's website, and registered the procedure in SEAP.
On Thursday morning, at 9:00, he walks into the mayor's office with the complete file.
The mayor looks at the stack of papers. — Done?
— Done.
— And it's legal?
— Legal.
The mayor signs without asking anything further.
What Actually Happened — The Anatomy of Three Layers
Now that the story has concluded, it is worth understanding precisely what happened on that Tuesday. Not because Gheorghe needs to understand — he is satisfied that the job is finished. But because we need to understand, if we want to talk honestly about digitalisation in public institutions.
Three layers were present, superimposed.
Layer one: Digitised manual work — what Gheorghe used to do. Word documents opened, copied, edited, printed, scanned, uploaded. This is not automation. It is hand work performed on a screen. It is the equivalent of writing with a pen on a tablet.
Layer two: Process automation — what the town hall could have done had it invested in RPA. A software robot that opens SEAP, fills in standard fields, saves documents in the correct locations, sends notifications. Fixed rules, rapid execution, zero judgement. It works perfectly for repetitive and identical tasks. It fails entirely when an exception arises, a novel situation appears, or a question must be answered that has no predefined response.
Layer three: Implemented AI — what Gheorghe used on Tuesday. Not execution, but reasoning. The platform read the context, selected the correct procedure from among multiple possibilities, generated documents adapted to the specific situation, checked them against an up-to-date legal database, and formulated clarifying questions where information was insufficient. This is not a robot. It is a system that thinks procedurally — within limits, within a defined domain, verifiably — but is genuinely capable of judgement.
The difference between layer two and layer three is not one of speed. It is one of nature. RPA executes steps it already knows. AI identifies the correct steps and adapts them to the situation.
Gheorghe's Moral Threshold
On the way home that evening, Gheorghe tries to put words to his discomfort.
It is not that he fears losing his job. Not yet. He knows that the town hall needs someone who can understand what the platform has produced. Without him, the mayor would have pressed Enter on a document with incorrect CPV codes and published an invitation with an illegal deadline and received a formal challenge within three days.
And yet. Nineteen years. Nineteen years in which he knew something that no one else in the town hall knew quite as well: how to produce a sound file. This knowledge was his professional identity. It was the reason why the mayor, however impatient, still had to come to him.
Now a machine reproduces that thing in forty-seven minutes.
He is not angry at the machine. He is disoriented by what remains. If documentation takes forty-seven minutes, then his nineteen years were not about documentation — they were about something else. About judgement. About local context. About knowing that a supplier from the city has an eight-week delivery lead time, whereas the one from the next town delivers in a fortnight but at lower quality. About relationships with ANAP. About how to draft a response to a formal challenge.
But he has not transmitted any of that to anyone. He never had time — he was busy with the technical specifications.
The Accountant and the Distribution of Knowledge
The following day, Gheorghe rings Mihai.
— Mihai, it worked for you. For you. But what if someone else had used it — the secretary, say, or a newcomer — and did not understand what the platform was producing?
— Then it would have gone wrong, says Mihai simply. Like any tool.
— So it's not for everyone.
— It's not for anyone who has never understood the procedure. It's perfect for someone who already understands the procedure and no longer wants to spend two days writing it from scratch.
That is the distinction Gheorghe carries home and turns over in his mind for several days.
The platform did not democratise public procurement. It did not make it accessible to anyone, regardless of competence. It compressed the drafting work for someone who already understands what they are drafting. For someone who does not understand, it produces plausible documents that can be disastrous.
There exists in this world a category of tools that amplifies competence, and another category that conceals incompetence. The AI procurement platform belongs to the first category — but it can be used as if it belonged to the second. And that difference is not made by the platform. It is made by the person.
The Replacement of Mindsets — A Harder Note
There is a third character in this story, invisible throughout. It is not Gheorghe, it is not Mihai. It is the civil servant at the neighbouring town hall — the one who received the same link from the same Mihai, with the same explanation, and closed it after thirty seconds with the words: "It's not for me. I have my own method."
His method produces correct files. Slowly, but correctly. He will not use the platform. Not from incompetence — from identity. He is his method. To abandon the method would be to admit that twenty years of practice were partly redundant. And that is not a conclusion one can bear on a Tuesday morning, with the smell of wet linden blossom coming in through the window.
The manager who wishes to implement AI systems in public institutions will encounter this person and say: "Resistance to change."
It is not resistance to change. It is adherence to identity.
And adherence to identity is not solved by a one-day training session or by a software licence. It resolves itself, over time, through one of two paths: either the person finds a new identity in which their old competence still has a place — as Gheorghe found; or the person retires, and their position is filled by someone who has never known any other way.
This is the mindset replacement that appears in every strategic document printed in Calibri 11. It is not a metaphor. It is literally what is happening in Tallinn, in Singapore, in Seattle: the people with the old mindset either transform — painfully, slowly, with something like embarrassment — or they are succeeded by people who have nothing to transform.
The difference between Romania and those countries is not that the people there are cleverer or braver. It is that there Mihai rang sooner, and the knowledge exchange in Cluj has been running for more years, and the mayor knew to ask for a complete file rather than simply for speed.
There is no third path. The platform does not adapt itself to the old Gheorghe. The old Gheorghe must become the new Gheorghe. And that costs considerably more than a software licence.
The Monday Morning After Thursday
On the 15th of July, the county delegation arrived. The council chamber was presentable — new chairs, a central table, a cabinet with sliding doors. No one mentioned the furniture. They spoke about county roads and European funding.
Gheorghe sat at the end of the table and took notes.
At one point, one of the county councillors asked: "And how are you getting on with digitalisation? There is European funding coming for digital public administration..."
The mayor looked at Gheorghe.
Gheorghe said: "We are getting on. More to do."
And he wrote something in his notebook.
"The transition from using an operating system to automating through a prompt does not happen by installing a programme. It happens in the moment when a person with nineteen years of experience sits for three minutes in front of a blank white text box and decides that they can trust it. That moment is not produced by any national digitalisation strategy. It is produced by a phone call from a friend who says: I did it, and it worked."
— marginal note, Gheorghe's notebook, July 2025
Critical Apparatus
Ten Questions and Answers
1. What is the real difference between "process automation" and "implemented AI"? Process automation (RPA) executes predefined steps according to fixed rules — the same actions, always in the same sequence, without judgement. Implemented AI reasons: it reads context, selects the correct procedure from among several options, adapts documents to the specific situation, and formulates questions when something is unclear. The difference is not one of speed. It is one of nature: execution versus procedural judgement.
2. Why is Estonia the most instructive comparison for post-communist administrations? Because Estonia began from a structurally comparable position — a post-communist administration, limited resources, a small population — and chose to build infrastructure before purchasing solutions. X-Road, the once-only principle, universal digital identity — none of these are AI products that were bought in. They are state architectures that were deliberately constructed. Many Eastern European governments purchase solutions without having built the architecture. The sequence matters enormously.
3. What is the difference between the "agentic AI" that Singapore is building and the generative AI that Gheorghe used? Gheorghe used generative AI: he described a situation, and the platform produced documents. Agentic AI is the next stage: a system that acts without being asked — monitors procurement deadlines, detects that a product code has changed in the legal database, automatically dispatches notifications, initiates procedures. Singapore is already moving in this direction: one agent per civil servant. This is not a chatbot. It is a colleague that never sleeps and never forgets a deadline.
4. Why is the civil servant at the neighbouring town hall not simply "resistant to change"? He is adherent to identity. His professional identity is built around a method. The demand that he abandon the method is experienced as a demand that he abandon his value. This is an identity conflict, not a competence problem. Managers who confuse the two deploy the wrong instruments. You do not solve an identity crisis with a training programme.
5. What are the ethical risks of using AI for procedural documentation in public administration? The principal risk is plausibility without accuracy: the platform produces documents that look correct but may contain subtle errors — an incorrectly updated product code, a deadline that does not conform to a recent legislative amendment, a technically unrealistic specification. Without a Gheorghe to verify, these errors enter the system bearing the authority of an official document. The accountability gap is real, and it is not the platform's problem to solve.
6. Why is it insufficient simply to purchase AI licences, as the British and American cases suggest? Because money solves the budget problem, not the mindset problem. The United Kingdom spent £573 million on government AI contracts and yet assessments reveal fragmentation and absence of internal capacity. Purchasing AI without building the capacity to evaluate and integrate it produces dependency, not competence. The gap between what is bought and what is understood is where public money disappears.
7. What did the town hall gain? Time (two days compressed into half a day), improved compliance (product codes verified automatically against the current catalogue), and a procedure opened more quickly — which increases the likelihood of competitive offers arriving. The gain is not spectacular. It is precisely what was needed for the mayor to sign on Thursday rather than the following Friday.
8. What did Gheorghe lose? An illusion: that his value resided in his capacity to produce documentation. He discovered that his value resides in his capacity to verify and contextualise — which is harder to replace, but also harder to claim as visible authority. The paradox is that he is more valuable now than he was before — but he feels less important. This is not unusual. It is, in fact, the defining emotional experience of the transition from craft to augmented practice.
9. What does "agentic team orchestration" mean for a small town hall? For the moment, nothing concrete. It is the horizon. Agentic orchestration means that multiple AI agents work in parallel on different tasks within a single process — one monitors the procurement platform for deadlines, another generates documents, another tracks supplier responses, another sends notifications. At a town hall like Gheorghe's, this remains speculative. But Estonia is building it now. Countries that delay will purchase the solution in 2035, when it will already have been superseded.
10. What must happen for public institutions to genuinely adopt AI? Not training programmes. Not software licences. Not national strategy documents. There must be more Mihais who ring more Gheorghes and say: I did it, and it worked. And the knowledge exchanges in Cluj — or wherever the local equivalent may be — must become systematic rather than accidental. The adoption of technology in public institutions is, at its core, a social phenomenon, not a technical one. It travels along the same networks as trust: through people who know each other, from football and from weddings.
Embedded Arguments (The Aesopic Layer)
In the Romanian literary and intellectual tradition, șopârle — literally "lizards" — refers to arguments embedded within apparently innocent text: a practice honed during the communist period, when direct speech was dangerous and indirection was survival. This article carries several.
Embedded argument 1:"Office furniture — chairs, desks, cupboards, shelving units" — the technical inventory mimics the banal. The actual subject of the narrative is the cognitive infrastructure of a bureaucratic system: what it stores, what it retrieves, what it can no longer protect.
Embedded argument 2:"It's not for me. I have my own method" — in the mouth of the invisible civil servant, a synecdoche for every institution that produces digitalisation reports without digitalising anything. Most countries have more of these institutions than they have Gheorghes.
Embedded argument 3:"The knowledge exchange in Cluj" — this is not a narrative detail for local colour. It is the only mechanism through which practical knowledge genuinely transfers within public administration: not through agency guidance documents, not through European webinars, but through Mihai ringing Gheorghe.
Embedded argument 4:"We are getting on. More to do." — the compressed response of a man who has just understood that digitalisation is not about platforms. It is about the willingness to put one's own identity in question. Six words. Nineteen years of subtext.
Embedded argument 5:The three columns on the screen in Cluj — process automation, implemented AI, digitised manual work — represent the only honest taxonomy of public sector digitalisation. What is missing, everywhere, is the institutional courage to apply this taxonomy to one's own desk before applying it to anyone else's.
Rhetorical Architecture
The piece operates through several overlapping structures:
Narrative within narrative — the knowledge exchange in Cluj, inserted as Mihai's reconstructed memory, provides a conceptual pause without breaking the fictional frame. The theoretical work is done by a character, not by the author.
The table as dramatic object — the three columns on the projector screen in Cluj function as the article's central conceptual clarification, but from within the story rather than above it. This is the difference between telling and showing, applied to an argument about technology.
The geographical character — the time zone, the border, the tractor outside the window anchor a global discussion (Estonia, Singapore, the UK) in the specific gravity of provincial Eastern Europe. The scale works in both directions: the universal made local, the local made universal.
The paradox of inverted value — Gheorghe is more professionally valuable after using the platform than he was before, but feels less important. This is the article's central conflict, formulated without being named. Naming it would diminish it.
The invisible third character — the civil servant at the neighbouring town hall who does not appear in any scene but who structures the moral conclusion. His absence from the narrative mirrors his resistance to change: he refuses to enter the story just as he refuses to enter the text box.
What You Can Do Tomorrow
If you are Gheorghe: Try the platform on a small document, one you know well. Not to save time — to test your own judgement. You will discover that you know more than you thought you did.
If you are the mayor: Do not demand Gheorghe's speed. Ask him to become a mentor to someone younger who will use the platform without understanding what they are verifying. Without Gheorghe transmitting his knowledge, the platform becomes a risk rather than an efficiency.
If you are Mihai: Ring another Gheorghe. And go to the knowledge exchange. Even if it means travelling.
If you are the manager who wants to implement AI institutionally: Abandon the idea that the problem is resistance. The problem is that you have not created the conditions in which people can discover they know more than they believed. Build the architecture before you buy the solutions. And do not confuse a national digitalisation strategy with a Gheorghe who received a useful link from a friend.
If you are a policymaker reading about Estonia: The lesson is not the technology. It is the sequence. Infrastructure first. Solutions second. People throughout.
If you are Estonia: Carry on.