
AI Is Not Always the Answer
95% of AI pilots deliver zero P&L impact. Technology doesn't fail — the approach does. Why asking 'what can AI do here?' is the wrong first question.
Field reports on AI experiments, innovation, insurance, search technology and day-to-day execution.

95% of AI pilots deliver zero P&L impact. Technology doesn't fail — the approach does. Why asking 'what can AI do here?' is the wrong first question.

Uber burned through its annual AI budget in four months. Microsoft pulled licences. Yet the problem is not AI itself — it is who watches the meter.

By 2029 there will be four AI winners, not one. Google takes the consumer market via data, chips, cash flow and distribution. Microsoft wins enterprise. Anthropic wins specialists. OpenAI stays a strong number two.

OpenAI joined C2PA and SynthID on May 19. My tests with two fake receipts show that neither OpenAI's verification tool nor Gemini detect the SynthID watermark from Google's own Gemini Nano Banana 2. What fraud departments should build instead.

Twelve out of thirteen AI models cited the deductible with the wrong year. One got it right: Gemini. Not because it's smarter, but because it has a better information stream. And that stream is called Google Search.

I build applications without a developer background. And every time a session runs long enough, something gets lost in the middle. Not through carelessness. Through architecture. The same effect that hits 160 million Eurovision viewers every year.

A year after my fake-receipt test, I ran the same prompt past ChatGPT, Codex and Gemini again. One company, two faces, and a new kind of AI safety you couldn't see a year ago.

Agentic commerce shifts the e-commerce battle from visibility to selection. What Google's UCP, ChatGPT shopping research and Bol's Billie mean for your webshop.

The dark patterns ban takes effect on June 19, 2026 with a mandatory cancel button for financial services websites. What does EU Directive 2023/2673 mean for your business?