NVIDIA Research releases HORIZON, an AI agent framework that evolves Verilog and RTL code using git worktrees

Original post

Cool new paper from NVIDIA.

Looks like agentic coding is moving into hardware design.

HORIZON treats hardware design as repository-level code evolution. A Markdown harness becomes a project pack with domain knowledge, an executable evaluator, an acceptance predicate, and a git policy.

The agent then evolves an isolated worktree.

That is a strong pattern because hardware design needs executable checks. The verifier harness becomes the real interface between the agent and the design task.

The paper reports 100% benchmark completion across several hardware design suites, which makes this one worth tracking even if you do not work on EDA.

Paper: https://arxiv.org/abs/2606.28279

Learn to build effective AI agents in our academy: https://academy.dair.ai/

5:11 PM · Jun 29, 2026 · 2.8K Views

NTK AI@NtokozoAI

The verifier harness is the part that travels.

Hardware design looks like a hard domain, but it has something most enterprise workflows do not: an executable way to say whether the work passed.

That is why the authors can report 100 percent benchmark completion and still be careful not to call the problem solved.

For agents, reliability will track the quality of the acceptance gate more than the difficulty of the domain.

6h5

BlakeTheCoder@kyr_dreamer

@RobertTLange I love the git worktree state management. 100% benchmark sweep feels too clean though. What's the compute cost per task? How's it handle timing closure?

17h2