AI Failures

"Developers are growing increasingly frustrated with AI coding tools that produce deceptively flawed solutions"

Three most likely explanations which most developers will dislike and reject.

AI Hype Cycle: the natural cycle of human perception, exuberance of expectation followed by the disappointment of reality. "It's all in your mind".

No Repeatable Process: Most developers can recite "repeatable process" but rarely understand rigorous routine. For instance, a developer choosing MacOS to develop Linux apps has already violated the concept of Repeatable Process with potential compatibility issues. A repeatable process counters the non-deterministic nature of generative AI.

Token Overload: "When developers push entire systems into a single prompt, they exceed the context window, triggering subtle, non‑deterministic errors that look plausible but break downstream logic. This isn’t intelligence failing; it’s complexity spilling over the edges... Treat AI more like a network with packet constraints—small, controlled exchanges lead to reliable outcomes, while oversized payloads silently drop information."

Almost all developers I've seen in the past fifteen years don't understand the reality of repeatable process. I was an SME for a potential lawsuit involving Verizon and an initial task was assessing their IT infrastructure and its level of repeatable process. File systems and configs of six host tiers were virtually identical with the exception of additional deployments on their PROD and PROD_TEST tiers. Verizon had military precision in their DEVOPS staff.

This Slashdot article supports my impression that younger (ie most) developers know a great deal about not very much. I identified the non-deterministic and token stack pitfalls after a couple of weeks of AI vibe coding.