Written in the open, and in progress. Live, evolving work that will keep changing. How this book is written →

Technical Building Blocks

Author
Affiliation

Harvard John A. Paulson School of Engineering and Applied Sciences

Published

August 11, 2026

Building reliable AI-native design loops requires a rigorous technical substrate. We cannot connect language models to hardware code generators and expect correct silicon. In Part II, we define four modular technical building blocks that every AI-native design loop needs: machine-usable hardware representations, learned design methods, tool-connected execution environments, and verification feedback.

We classify the structural representations that make hardware design state machine-usable, connect learned methods to cycle-accurate simulators and physical-design tools, and construct the verification feedback that detects and contains proxy gaming.

Part overview
  1. How do we maintain machine-usable state alignment between abstract system models and physical silicon constraints?
  2. What qualification rules prevent unverified candidate proposals from corrupting subsequent design search steps?

Chapter Roadmap

Chapter Core Architecture Question Key Takeaway / Deliverable
4  Architecture Data, Knowledge, and Representation Which structural representations preserve hardware constraints for AI models? Classifies structural representations (syntax trees, dataflow graphs, transaction-level models), data provenance, and access-bounded process knowledge.
5  Prediction, Generation, and Optimization How do prediction, generation, and optimization methods compare across architectural tasks? Builds the role, method-family, and technique selection framework with the smallest-sufficient-approach rule.
6  Execution Harnesses and Tool Isolation How do we build scalable interfaces between AI loops and complex CAD toolchains? Builds the tool, wrapper, harness, and environment split with pinned state, typed contracts, and execution records.
7  Verification, Feedback, and Learning How do we turn simulation and timing errors into verification feedback that drives search? Establishes the qualify-then-claim-then-act sequence, proxy diagnostics, and independent checks.