AI failures do not always appear as dramatic breakdowns. They often emerge as inconsistency, partial success, and decisions that become difficult to explain after complex systems have acted across many steps.
In “The ‘Hot Mess’ and the New Means of Production,” I examine what happens when this uncertainty moves upstream into the environments where software is created. Agentic AI systems can now plan tasks, generate and revise code, call tools, and participate in architectural decisions with limited human intervention.
Apple, Microsoft, and other major platforms increasingly control the development environments, model selections, workflows, and defaults through which AI-assisted software is produced. Those choices travel downstream. Public agencies, schools, hospitals, nonprofits, and businesses often procure systems built inside production environments they did not design and cannot easily inspect.
This makes AI governance a supply-chain issue. Institutions inherit more than a finished product. They inherit platform assumptions, model behavior, delegated complexity, and decisions about what gets logged, retained, automated, or made visible to human oversight.
Governance must therefore begin before deployment. Institutions need stronger vendor disclosure, traceable development records, resilience planning, and exit strategies for the platforms and AI systems that have become critical production infrastructure.
Read the complete essay, “The ‘Hot Mess’ and the New Means of Production,” in The Ready Futures Journal on Medium.