Top story 1NVIDIA has introduced Auto Ontology, an open Python framework that builds graph-based semantic reasoning layers over enterprise data and exposes them to agents via MCP, CLI, and API. The framework targets a core reliability problem in agentic workflows: grounding context in discoverable, explainable data rather than ad hoc retrieval. If adopted broadly, it could standardize how enterprise agents reason over governed knowledge graphs.
G GitHubTop story 2TSMC presented a BEOL-integrated 1T1C oxide-semiconductor memory at the 2025 VLSI Symposium, reporting 0.75V operation, over 128ms retention, sub-8ns read/write cycles, and endurance above 10¹⁴ cycles. The technology could offer denser on-die cache than SRAM, a meaningful lever for future AI and advanced-logic chips. It remains a test-chip result, but the cache-density implications are significant for compute scaling.
r RedditTop story 3A new Harness Leaderboard post benchmarks coding agents and harnesses including Kimi Code, Hermes, Pi, Claude Code, and Codex on Terminal Bench 2.1 and Deep SWE 1.1. The comparison gives practitioners a useful view of agent performance beyond raw model benchmarks. Harness design is emerging as a major variable in real-world coding-agent productivity.
r Reddit