From 'I'm an engineer' to 'ChatGPT says so': AI now balances the books
AI already reviews financial statements, generates entries for Spain's models 303 and 347, and spots discrepancies at the level of A3. Technical authority no longer comes from a professional degree.
English · Original discussion in Spanish · Published
From 'I'm an engineer' to 'ChatGPT says so': AI now balances tax accounts
The shift in authority is the least discussed economic story of the year. Technical arguments are no longer settled by appealing to a degree —'I have qualifications, I'm an engineer'— but to the machine: 'ChatGPT says so' now occupies the place once held by 'it was on TV'. This is not a change of habits. It is a change on the balance sheet. Behind it lies a concrete reorganisation of which jobs still get paid and which are starting to become redundant.
AI that already handles the 303 and the 347
With a portfolio of over a thousand clients of all kinds, it feels like the ground has shifted in a matter of months. AI tools review tax accounts, profit and loss statements and invoice ledgers, and spot discrepancies that used to take hours to find. The next step has already been taken: entries for forms such as the 303, the 340 or the 347 are now generated with the help of Claude, ChatGPT and local software. And in some cases the result is on a par with A3, the platform that has been the standard in Spanish accounting firms for decades.
That is the uncomfortable truth. When a general-purpose tool matches specialised, well-established software, the marginal cost of a bookkeeping entry approaches zero. The breakdown by tax forms and whatever still falls outside automation is the part that cannot be settled in a single sentence.
The controller becomes a hybrid
Something similar is happening in the finance function. The classic controller role is losing weight, and everything points to a hybrid survivor: someone who no longer competes with the machine at consolidating data, but interprets what the machine cannot see. The same logic is emerging in technical arguments before the AE (Spain's tax agency), where firm work is starting to look less like writing and more like verifying.
The student who goes blank
The phenomenon does not stop at the workplace. At university, law assignments come back answered by AI: correct as long as the script is trinc, silent the moment the student is taken off the syllabus. The pattern is identical to the office one. AI solves what is routine and fails at what is not — exactly what an oral exam measures, and exactly what a tax inspection measures.
SSMS credentials over WhatsApp
The most revealing case is that of a veteran engineer at a top-tier company: cock of the walk in technical meetings, and then out-of-channel requests —the backup, the SSMS credentials— over private messaging, with infrastructure management, VPN and CheckPoint involved. The problem is not AI. It is the incentive: zero ambition, early retirement on the horizon.
The failure nobody audits
The obvious sustancia ilegal remains. ChatGPT will not admit it doesn't know: if it lacks the data, it invents it with a confidence that sounds like a report. For anything specific you need to rely on results with verifiable links and judge the source afterwards. The bottleneck is here: when the machine gets an entry wrong, nobody has yet decided who signs off on the error.
The shift in authority is the least discussed economic story of the year. Technical arguments are no longer settled by appealing to a degree —'I have qualifications, I'm an engineer'— but to the machine: 'ChatGPT says so' now occupies the place once held by 'it was on TV'. This is not a change of habits. It is a change on the balance sheet. Behind it lies a concrete reorganisation of which jobs still get paid and which are starting to become redundant.
AI that already handles the 303 and the 347
With a portfolio of over a thousand clients of all kinds, it feels like the ground has shifted in a matter of months. AI tools review tax accounts, profit and loss statements and invoice ledgers, and spot discrepancies that used to take hours to find. The next step has already been taken: entries for forms such as the 303, the 340 or the 347 are now generated with the help of Claude, ChatGPT and local software. And in some cases the result is on a par with A3, the platform that has been the standard in Spanish accounting firms for decades.
That is the uncomfortable truth. When a general-purpose tool matches specialised, well-established software, the marginal cost of a bookkeeping entry approaches zero. The breakdown by tax forms and whatever still falls outside automation is the part that cannot be settled in a single sentence.
The controller becomes a hybrid
Something similar is happening in the finance function. The classic controller role is losing weight, and everything points to a hybrid survivor: someone who no longer competes with the machine at consolidating data, but interprets what the machine cannot see. The same logic is emerging in technical arguments before the AE (Spain's tax agency), where firm work is starting to look less like writing and more like verifying.
The student who goes blank
The phenomenon does not stop at the workplace. At university, law assignments come back answered by AI: correct as long as the script is trinc, silent the moment the student is taken off the syllabus. The pattern is identical to the office one. AI solves what is routine and fails at what is not — exactly what an oral exam measures, and exactly what a tax inspection measures.
SSMS credentials over WhatsApp
The most revealing case is that of a veteran engineer at a top-tier company: cock of the walk in technical meetings, and then out-of-channel requests —the backup, the SSMS credentials— over private messaging, with infrastructure management, VPN and CheckPoint involved. The problem is not AI. It is the incentive: zero ambition, early retirement on the horizon.
The failure nobody audits
The obvious sustancia ilegal remains. ChatGPT will not admit it doesn't know: if it lacks the data, it invents it with a confidence that sounds like a report. For anything specific you need to rely on results with verifiable links and judge the source afterwards. The bottleneck is here: when the machine gets an entry wrong, nobody has yet decided who signs off on the error.
Also available in: Deutsch
Summary of a discussion on Burbuja.info - Foro de economía, actualidad y política., translated from Spanish and reviewed before publication. Read the full discussion (18 replies).
More summaries
-
ADE or Law: Which Degree Doesn't Guarantee a Decent Salary
ADE vs Law: a double graduate claims only 0% to 10% of practicing lawyers live well
-
Couscous vending machines near migrant youth centers: business or bust?
An analysis of the viability of placing couscous vending machines next to centers for unaccompanied migrant minors: margins, real demand and operational risks debunk the myth.
-
Esquelas Cantabria: funeral death notices startup hunts for a business model
A death notices portal for Cantabria, Spain, has sparked debate over its viability: advertising, subscriptions or charging funeral homes. Does it have a future?
-
Companion services website sells for 700 euros
A website with growing organic traffic and 200-euro sponsored posts is listed for 700. The price clashes with legal doubts and valuation concerns.
-
Why an entrepreneur picks free QGIS over paid ArcGIS
An entrepreneur uses free QGIS for a landscape impact study for the Catalan government and saves the ArcGIS licence fee. Free software versus bureaucracy.
-
Free Canva alternatives: the fine print of 'free'
Free alternatives to Canva: Microsoft Designer with no limits, LibreOffice Draw, GIMP and Inkscape with a learning curve. Free always comes with a catch.