Who Gets to Move the Recipe? How Autonomous Agents Earn Decision Authority in the Fab

What happened

This is an opinion/analysis article (not a procurement-relevant operational event) by data and AI architect Ajay Kumar Govindaram on how autonomous agents earn decision authority in a semiconductor fab. It proposes a four-level autonomy ladder (observe, recommend, act with approval, act within limits), with authority granted per decision class and promotion gated by shadow-mode agreement, a minimum sample size, and an SPC control chart on the agent's own conduct; demotion is automatic when the agent's behavior metrics cross a control limit. The article cites Gartner's June 2025 prediction that over 40% of agentic AI projects will be canceled by the end of 2027, pointing to escalating costs, unclear business value, and inadequate risk controls. It also mentions that PDF Solutions introduced Exensio Aurora in September 2026, a platform built from the start to run agentic AI across manufacturing operations at petabyte scale on a contextualized, trustworthy data layer. The article notes that older 200mm or mixed brownfield lines often lack tool connectivity and data quality needed to run SPC on an agent, capping that agent's autonomy. The views expressed are the author's own, with illustrative or publicly sourced examples.

Impact and signal

Affected components and suppliers

Source and time

Source: SemiWiki · 10/20/2026 · View original source