Humans vs. vibe-coded agents or, a fun little obituary for classic consulting
More and more people are using AI — and so are the people who hand out loans and finance projects.
So let's try to peek into 2030 2027. It feels like the familiar PDF business plan, or a market research deck with smiling target customers, will become an anachronism — like a Soviet savings passbook.
Right now that document is a ritual offering to the bank or the investor. The entrepreneur (or, more often, the economist they hired — that's literally me, 14 honest years of this grind 🧐) doesn't sleep for a week, torturing Google and inventing fairy tales about market domination, while the bank's credit officer spends a week pretending to read it (also me, back in my 10 years at Sberbank).
So what's coming?
Some AI service (say, my PlanMaster, or a new skill from Anthropic) gets a voice message from the businessman — and then the service doesn't write prose. It packs the data into a JSON file containing: - competitors' receipts from fiscal data; - foot traffic near the hypothetical café; - and a mathematical model of whether the business survives 20% inflation and the next war with Iran.
The agent knocks on the bank's API, where an equally vibe-coded agent receives the package, runs it through 100,000 simulations in nanoseconds, and pushes the credit department head a QR code to issue the entrepreneur their loan. That's if it's a yes. If it's a no, it sends the package back with a TODO list of what to fix 😂😂
So why do we even need humans? 🫤
Well, first, you need at least one lunatic with an economics degree on each side. Someone has to write those agents, after all.
Second, someone absolutely has to knock back a whiskey to celebrate the bank's beautiful numbers — a bank now staffed by a director, a prompt engineer, and… a cleaning lady.
You probably also need a person to verify the collateral pledged to the bank. Though, remembering how I froze in cold-storage rooms inspecting 40 tons of frozen herring held as collateral, that job is better handed to a drone.
Everyone else — economists, marketers, inspectors, and assorted clerk-walkers — sip instant coffee and read news on Telegram about the automation of credit and other operations.
Crank the cynicism up to max, and the picture looks like this: a) The entrepreneur buys a subscription to a web service (I'm building it as hard as I can 😵💫) that packs even the most hopeless case into JSON so it looks like the next Google. b) The bank fires its army of inspectors, and the one remaining programmer plus Claude write an Inquisitor Agent, trained to catch the micro-embellishments in application data. c) Then two servers fiercely argue with each other and swap terabytes of data at the speed of light. d) Meanwhile the entrepreneur and the banker sit in a bar planning a fishing trip over a beer. Nothing depends on them anymore anyway. Just renew the subscription..
Yes, obviously there are legal nuances — you'd have to rewrite the bank's charter. Again, with the help of Claude or Codex)) The big deals will remain, where the cost of an error is too high and no one will hand responsibility to an algorithm. But what percentage of deals is that?
I'm genuinely curious when some N-bank will be the first to admit that its entire credit committee is just a Python script.
The main takeaway: do your consulting with the neural network's later processing in mind, too.
#upgrowplan #ArtificialIntelligence #DataQuality