The Workflow Layer: Where AI Can Change Semiconductor Engineering
The bottleneck is not generating more engineering output. It is shortening the path from new evidence to a decision engineers can trust.
Long-form writing
Essays
Analysis of AI, engineering workflows, and the systems required to connect them.
-
-
The Abundance Effect: Why AI Efficiency Can Increase Demand and Move Bottlenecks
When AI makes a task cheaper, more of that task becomes worth doing. Efficiency can expand demand and move the bottleneck instead of removing it.
-
AI at System Scale: The Control Plane Economy
As intelligence becomes abundant, coordination becomes scarce. Value moves to the layers that connect state, policy, action, and feedback.
-
Compiling Enterprise Knowledge for Reliable Agent Systems
Enterprise knowledge becomes useful to agents only when it is compiled into context, tools, rules, evaluators, and approval paths.
-
When Output Becomes Cheap: Where Human Value Moves
AI can make output abundant without making judgment, understanding, or accountability abundant. Human value moves toward what remains scarce.
-
Why Complex Domains Need Verifiable Learning Loops
AI can search far beyond human scale only when the surrounding system can cheaply and reliably reject wrong answers.
-
Build the Domain Layer, Not Another Agent Runtime
Generic runtimes are becoming infrastructure. Durable advantage lies in the domain layer: context, tools, permissions, evaluation, and operational memory.
-
Beyond Transformer and Mamba: A Case for Layered Memory
No single mechanism can preserve exact history and compress endless streams at low cost. Intelligent systems need different kinds of memory for different timescales.
-
Turning Agent Experience into Durable Memory
Logs preserve what happened. Durable agent memory requires a governed process for selecting, testing, consolidating, and retiring experience.