There’s a word MIT hackers coined in the 1950s: cruft — useless, tangled, accumulated junk that makes a system incomprehensible and impossible to build on.

Sound familiar?

In 2026, we have a new variant: crufty AI.

Chatbots bolted onto data silos. LLMs fed dirty, unstructured, undocumented inputs. Automation layered on top of processes nobody fully understands anymore. Pilots that never graduate. Dashboards nobody uses. Vendors paid. ROI: zero.

This isn’t an AI problem. It’s a cruft problem.

You can’t get intelligence out of a system that was never coherent to begin with. The model isn’t the issue — the substrate is.

Your AI isn’t underperforming. Your architecture is.

Before you buy another AI license, ask:

  • Is your data governed, or just stored?
  • Do your systems integrate, or just coexist?
  • Does anyone have a current map of what actually runs this business?

AI amplifies what’s already there. If what’s already there is crufty, you’ll get crufty results — faster and at greater expense.

The fix isn’t a better model. It’s architecture.