Blog

Booking.com's calendar feed marks a real reservation and a manually blocked date with the identical string. Reading the feed was trivial; knowing what it meant took a second signal the feed never carries — and the same ambiguity walked back in on three surfaces. The semantic gap in every enterprise integration.

A retired €0 phone runs an always-on AI agent that watches my bookings with no computer attached. The model was the easy part — surviving a hostile OS, authenticating without a human, and recovering after reboot is the forward-deployed last mile of agentic AI.

A small business spent €15 a day on ads and the platform reported zero conversions — while real bookings kept arriving. The algorithm was optimizing on the wrong number. Closing that gap is forward-deployed work, even for a three-apartment rental.

Budgets approved, demos impressive, production empty. The enterprise AI failure is almost never the model — it is the last mile. The numbers, the five root causes, and what the 12% that ship do differently.

Every vendor wants to sell you an AI platform. You probably don't need one. Capable AI is already on the market — what's missing is the wiring into your real systems.

In 2026 "Forward Deployed Engineer" went from a niche term to one of the hottest roles in tech. What the role is, why it exploded, and the 90-day shape of an engagement.

We replaced manual PR review with a swarm of specialized agents — a security scanner, a logic checker, a style enforcer, a test analyzer, and an orchestrator. What we learned in production.

Anthropic's official Telegram integration doesn't work on headless Linux. We built a workaround with ccgram and hooks on a free Oracle ARM VM. The full architecture.