South Korea's "AI for All" hands 52 million people free, unlimited AI — technical permission at national scale. Part one's evidence says that is half the job. Korea's own baseline (60% conceal), Romania's ION, and three indicators to watch.
Three weeks after Google, the DfT and NATS launched an AI programme for the North Atlantic, a NATS flight-processing fault helped cancel 2,000+ flights. They're not connected — and that's the point. Third failure in three years, one closed file, and the layer missing from every travel-risk model.
„Did you write this with AI?" The question doesn’t test the truth, but the author’s legitimacy. 57% hide their AI use. The problem isn’t adoption — it’s permission, technical vs. social. Plus the double trust penalty, human–AI.
The series' synthesis piece: one person, four unrelated domains, four products on the OpenAI platform 12 days after the catalog launched — all four published by September. The real chronology, the method that repeated, what didn't scale and what I standardized. Every claim verifiable with one click.
The series' final piece changes the perspective: 900 million users, an app catalog and Presence — software is moving into the conversation. What four shipped applications prove about how AI adoption actually gets done, and the questions I'd ask a product team on Monday morning.
Case study
The last technical study in the series, written with deliberate restraint: capabilities, not internals. 54 sources across 11 risk domains, monitoring as state and explainable silence — how a disruption radar becomes one judgment on your route.
Case study
Third study in the series: how to turn scattered government procedures into a conversational assistant. 270 documents in 9 languages, per-country reporting channels, and anti-hallucination as architecture, not as a filter.
Case study
Second study in the series: a TV guide for the agentic web. Tool selection as product design, verifiable freshness for data that expires by the hour, and the three-wave evolution from "what's on TV?" to "you choose for me".
Case study
How I turned Bucharest's waste-collection schedules into an application you can ask directly in ChatGPT: the problem, the architecture, the technical decisions and the lessons — with a diagram and verifiable facts only.
On 9 July, OpenAI opened its plugin platform. On 21 July, my product was live in it. Twelve days, one person, a personal project built in free time — and what that demonstrates about the difference between talking about AI and delivering it.
Most IT budgets miss their cheapest line item: someone on the client side who can read a technical estimate and tell you which lines are real and which are padding. Everything else gets more expensive because of it.
The pilot lasted three months. The fine — €14.4 million. Spain's DPA fined Amadeus for using PNR data for a purpose nobody had authorized. A purpose-limitation lesson for every TMC.
Google launched the Universal Commerce Protocol and is extending it to hotel booking. Half the industry writes OTA obituaries, the other half shrugs. Both look the wrong way.
In 2026, non-decision on AI adoption is the most expensive call a TMC can make. Why the EU AI Act is your quality filter, not your brake — and what you do Monday morning.
The EU AI Act brings concrete obligations for Romanian companies: fixed deadlines, mandatory AI literacy, and fines up to EUR 35 million. What it means for TMCs and the tourism industry.
How I built an interactive platform that transforms raw Eurostat data into actionable insights about European tourism across 27 countries.