In the AI feed, the world has already flipped upside down.
Old companies are about to collapse. Jobs will vanish. If you haven’t jumped on AI today, you’re irrelevant tomorrow. Any delay looks like falling behind forever.
I understand this feeling. I’m probably more inside this bubble than outside it.
I’ve been using top-tier subscriptions to ChatGPT, Claude, Chinese models, and local models for several years. In our landscaping business, we set up a dedicated service and embedded AI into visualization, documentation, and workflows. We ran local models on our own hardware. We built projects with AI under the hood: incoming messages, corporate knowledge bases, tender documentation, as-built documentation in construction.
So I’m not arguing against AI from a “tried it once, didn’t like it” standpoint. Quite the opposite. We use it heavily.
That’s precisely why I don’t buy the simplistic version of the hype.
AI is already changing business. Just not the way it looks from inside the bubble.
Business Doesn’t Move at Browser-Tab Speed
In personal work, a powerful model delivers a quick wow effect. It writes, explains, drafts, spots errors. Yesterday you did it yourself; today you’ve got a smart assistant by your side.
But a company doesn’t buy the feeling of a leap. It needs repeatable results: on its own data, within its own process, with a clear cost of error, an accountable person, and economics you can actually calculate.
A good AI system is closer to hiring an employee than buying a subscription. It needs access, tools, rules, oversight, quality criteria. Without those, it remains a smart browser tab. The tab is impressive. But it hasn’t yet changed the unit economics of the business.
Over the past month, I’ve had five conversations with companies generating over a billion rubles in revenue and about ten conversations with companies above 100 million rubles. This isn’t market research — just my working sample. But it’s sobering after scrolling through the AI feed.
Almost everyone knows about AI. Many understand the potential. But the conversations rarely head toward “let’s replace everyone tomorrow.” They quickly turn boring and practical:
- where exactly to apply it;
- what data is available;
- who will verify the output;
- what an error costs;
- how to measure the impact;
- who inside the company will own it.
This isn’t business being slow. It’s a normal reaction to a tool that needs to work in a live process, not a demo.
The Bubble’s Biggest Mistake
People inside the AI bubble often project their own adoption speed onto the entire market.
If I’m already living with agents, then business must be about to restructure just as fast. If I built a prototype in an evening, then a legacy company with a thousand employees should retool its process just as quickly by tomorrow. If a model can produce a draft, then a department can be downsized almost immediately.
But business doesn’t move like that.
It has data, access controls, security, regulations, legacy systems, habits, accountability, and the cost of error. None of that disappears just because the model got smarter.
AI can dramatically speed up a single step. But a company isn’t buying a faster step — it’s buying an outcome: lower costs, higher speed, fewer errors, more revenue, clear accountability.
Without that, it doesn’t matter how elegantly the model reasons.
Where Real Impact Appears
I don’t see the main potential in selling “AI” as a standalone miracle.
There’s a stronger scenario: a company enters an existing market and uses AI under the hood. It doesn’t make a religion out of it. It doesn’t try to convince the client to believe in neural networks. It simply changes the economics of the work.
Responds to inquiries faster. Processes documents cheaper. Finds relevant tenders better. Assembles commercial proposals faster. Needs fewer people for data entry and cross-checking. Delivers the same result to the client, but faster, cheaper, or more consistently.
That’s what becomes dangerous for incumbent players.
Not because ChatGPT will replace them tomorrow. But because a competitor will appear next to them selling the same result, but assembling it differently.
One Example: As-Built Documentation
As-built documentation in construction is, for me, a good illustration of exactly this kind of shift. Not the main thesis — just a concrete example.
From the outside, the topic sounds narrow: inspection reports, registries, quality documents, document sets, approvals. But on a large industrial construction site, this is a distinct production line. In Russia, there are companies with billion-ruble revenues that specialize in preparing as-built documentation. They may employ hundreds of people. Every day they collect, cross-check, transfer, verify, and format documents.
AI can’t simply “do the engineer’s job.” That phrase means nothing.
But it can take over specific pieces of work: find the right quality documents, verify completeness, prepare a draft inspection report, compile a registry, highlight discrepancies, show which data is missing.
The engineer doesn’t disappear. They review, correct, take responsibility, and sign off on the final result. If the system is built properly, you can see what the model suggested, what the person changed, and where errors most often occur.
From the outside, the result is the same: a set of as-built documentation. Inside, the mechanics are different: less manual assembly, faster verification, fewer snags, lower processing cost per set.
That’s how AI becomes a business tool. Not through a slogan, but by changing the unit cost of specific work.
Why “Adopt AI Urgently” Is Bad Advice
The phrase “we need to urgently adopt AI” sounds energetic, but it usually lacks an object. Adopt where? Into which process? At what cost of error? Who will be accountable for the result? What counts as success?
Giving every employee a ChatGPT or Claude subscription is useful. It can boost personal productivity, help strong performers work faster, and give the team experience. But it’s not yet enterprise adoption.
Enterprise adoption starts where there’s a workflow and clear economics:
- incoming inquiries that need fast triage;
- documents that need verification;
- tenders that need finding and evaluating;
- a knowledge base where employees waste time searching;
- as-built documentation where an error shifts timelines and payments.
That’s where you can start discussing the process. What AI does. What the person checks. Where the audit trail lives. How you count the savings. What happens if the model gets it wrong.
Without this, AI easily turns into a showcase: impressive to demo, hard to integrate, impossible to measure.
Where It’s Easy to Fool Yourself
There are two equally bad reactions.
The first: believe the hype and decide the model will replace everything right now. It won’t. Data is messy, context is complex, verification remains necessary, operations cost money, and people are still accountable for the outcome.
The second: dismiss it because “it’s not perfect yet.” That’s also a mistake. AI doesn’t need to be perfect to change the economics of individual operations. It’s enough for it to take a noticeable chunk out of manual preparation, searching, cross-checking, and grunt work.
While some debate whether AI can be trusted with the entire end-to-end process, others are already taking the pieces where the impact is measurable.
Instead of a Conclusion
The bubble isn’t wrong that AI matters. It does.
The mistake is elsewhere: expecting business to change like an X or Telegram feed. Fast, loud, with one big announcement.
In reality, much of it will be quieter. An inquiry got processed faster. A document was found sooner. A registry was assembled without manual busywork. A tender was filtered out before a manager spent half a day on it. A client got a reply faster than the competitor.
From the outside, the company seems to be doing the same thing. Inside, the cost structure has changed.
And that’s more dangerous for incumbents than loud predictions. A prediction can be ignored. A new cost structure operating right next to you — not so much.