ARI was born around a table, from a conversation between Federico and Daniele about what AI could do, and especially what it couldn't. It was the era when models were still clumsy: GPT-3.5 didn't exist yet, and every response felt surprising or wrong, rarely reliable. The intuition was counterintuitive: instead of asking AI to do more, ask it to do less. Only the parts that traditional programming can't reach. Everything else: deterministic code, testable, repeatable.

The first wow moment came with a small script (the POC of what would become ARIdoc) that took the photo of a receipt and returned a clean JSON, ready to integrate. A single example, but enough to convince us the direction was right: build a framework that gives statistical confidence on the "uncertain" piece, and leaves classical code to handle everything else.

From that embryonic idea ARI has grown a lot. Today it orchestrates complex workflows, integrates LLMs, RAG, MCP, agents, but the guiding principle has stayed the same: AI does the hard part, software does the reliable part.

The name comes from Ariadne. In the Greek myth she doesn't fight the Minotaur: she gives Theseus the thread to find his way out of the labyrinth. In a field where AI can feel like a maze of promises, complexity and black boxes, ARI is that thread: the trail that lets you get to production without getting lost.