How did I end up in AI? And why do I work only with back-office processes?
I'm a financier by education. Back in 1999, I went to study economics to understand how business really works. Over years in consulting, my team and I built around a thousand financial and business models. We worked as subcontractors for major consulting firms, serving large banks and clients from agricultural holdings to manufacturers. By 2023, with 16 years of consulting behind me, I had a solid understanding of business from the inside: what modules it consists of, where money flows, where the bottlenecks are.
That's when Sergey Kobelev pulled me into AI. He had just left corporate to focus on AI implementation. As I immersed myself, I realized AI is a fundamental technology. I was so impressed by its capabilities that my life split into before and after. I dove headfirst into a completely unfamiliar profession — one that didn't even have a name yet.
My first encounter with reality was simple. Before AI, my team and I handled the financial and market sections of Due Diligence reports. A 180-200 page report used to require 3-5 analysts working for at least 2 months — but AI changed everything. I started running individual sections through ChatGPT, then entire reports. First, it took a week solo. Now — hours, sometimes minutes, and the work of my team is done by AI research agents.
But classic AI in chat interfaces is poorly suited for business: hallucinations, instability, different answers to the same query. So I began constraining the model within strict boundaries. First a few rules, then system prompts, then I moved part of the logic into separate files — that's how the knowledge base emerged. When even that started hallucinating when accessing files, I arrived at semantic search. Then I moved from chat providers to API and CLI to keep control over output quality and token economics. That's how my own architecture emerged — one I now adapt to each client's specific challenges.
Why the back office?
Because business processes run through everything. You don't need to choose between sales and marketing. With the methodology in hand, you can find a bottleneck or an AI-ready task in any function. The architect here is not a developer or an implementer — it's someone who sees the full picture and assembles a solution with predictable results.
The big shift happened when I built the corporate agent factory — now called Tsekh 4.0. It became clear that you no longer need to learn to be a developer; you just need to frame the task correctly, and the solution creates the agent for it.
Another insight: it's wrong to measure AI impact by saved hours and money. The symbiosis of human and AI opens access to solving problems of a scale that simply didn't fit into your reality before.
AI delivers results when we keep the thinking and orchestration for ourselves and hand over the routine. Hand over the thinking too — stop reading, doubting, verifying — and we'll end up in a state that's harder to climb out of than thinking for yourself.