Case studies, breakdowns and field notes. No invented metrics or marketing gloss — only what has actually been validated on my own and client projects.
I designed and built the anxiety bot "Ministry of Existential Safety" end to end. The decisive move wasn't "wrap a model around everything" — it was drawing a boundary: the model writes only the satire, while the therapy and the crisis path run on deterministic code with no model call at all. The boundary is enforced in code, not in a prompt.
A Telegram bot built with AI, shipped to production, and still alive. The case covers blind mutual reveal as an invariant, splitting clean logic from the Telegram shell, a bug caught by hand in the live database, and why the real work starts AFTER launch.
The finale of my breakdown of AI content effectiveness. The machine is already infrastructure: $3.70 per dollar, 11–13 hours a week, 67% of businesses on it. But position 1 in search, the high-stakes deal and brand voice it can't carry. So the hybrid wins: the conveyor to the machine, trust to the human. How to build it and not mix up the shelves — on the Cadbury and BuzzFeed cases.
Part 2 of my breakdown of AI content effectiveness — on distribution. Google doesn't ban AI: 86.5% of top-ranking pages contain it, and the correlation between AI share and position is zero (0.011). But a human wins position 1 in 80.5% of cases vs ~10% for pure AI. And in paid ads AI matches humans on clicks — yet the "made by AI" label costs conversion. Only verified sources: Ahrefs, Semrush, Taboola.
Part 1 of my breakdown of AI content effectiveness. Blind, people can't tell machine text apart (53% accuracy) and even pick it as more engaging (56%). But admit "this is AI" and trust collapses: 52% disengage on the spot, and "AI-generated" scores nearly half the trust of "AI-assisted." Built on primary sources.
A client case (under NDA): I reconciled 1,345 CRM leads from a private Dubai clinic against their real WhatsApp chats. The clinic was booking roughly 1.5× more patients than its system showed, and ~740 conversations never entered the CRM at all. A step-by-step look at where revenue leaks — and why the report, not the pitch, is the product.
A client case (under NDA): in 8 weeks I re-architected the commercial model of an established player in professional education. I pulled raw sales data, found a hidden loss on off-site events, moved the business to subscription, and designed an AI agent for upsell. Forecast: −€43,000/year in costs, +72% LTV, +8–12% revenue.
Part 2 of my read on 2026: how AI slipped underneath ordinary life. AI-tutored students beat the control group by 0.15 SD, Apple now pays Google ~$1B/year to run Siri on Gemini, India's AI TB screening lifted detection ~16% — and a 12.1-point North–South divide widened. A sourced look at learning, health, the smart home, and gaming.
ChatGPT reached 1 billion users in ~3 years, and 2026 became the year AI moved from chatbots to agentic systems. A first-person, sourced read: how fast AI spread, where it landed, and how it rewired manufacturing, service, and strategy. Part 1 of 2.
Organic CTR on AI-Overview queries collapsed 61% — while the AI-search visitor became 4.4× more valuable. A sourced breakdown of the core contradiction of 2025: what happened to the funnel, who has already won and lost, and why strong SEO no longer carries you into the AI answer.
I built my own site in AI-native development mode — designing the code and the positioning at the same time, in one conversation with AI. This case covers why the offer had to be born together with the site, three artifacts that prove the method, and why the site is a Proof of Concept.
I advise businesses on revenue growth through AI — and yet ChatGPT, Perplexity and Gemini couldn't find my own site. The cobbler with no shoes. This case study walks through what I did: pre-rendering, opening the door to AI bots, Schema.org, llms.txt, hreflang, speed, IndexNow.
More notes coming soon.