• Research: Enterprise AI Workloads Are Tipping Toward Private Cloud — Broadcom's 2026 Private Cloud Outlook report finds production AI inference is shifting decisively from public to private cloud, with 56% of enterprises running or planning to run AI inferencing on private infrastructure. 🔗 Graph: TritonAI, UC San Diego, San Diego Supercomputer Center, Enterprise Monitoring 📅 Published: 2026-07-01 📰 https://campustechnology.com/articles/2026/07/01/research-enterprise-ai-workloads-are-tipping-toward-private-cloud.aspx 📌 Key takeaways: • 56% of enterprises surveyed are running or planning production AI inference on private cloud; public cloud for the same workloads dropped 15 percentage points year-over-year (56% → 41%) • 83% of enterprises are considering or have already repatriated workloads from public to private cloud — up from 69% in 2025 — and 43% are moving AI training, LLMs, and inference specifically • Cost management overtook security as the top public cloud challenge (31% of respondents); 97% believe some portion of public cloud spend is wasted • Directly validates TritonAI's on-prem strategy at SDSC: running sensitive AI workloads and inference inside the institutional firewall is becoming the enterprise norm, not the exception

• ScarfBench: Benchmarking AI Agents for Enterprise Java Framework Migration — IBM Research releases an open benchmark revealing that frontier coding agents are overconfident and struggle with the configuration complexity of real-world enterprise migrations. 🔗 Graph: Agentic AI, Model Context Protocol, Claude Code 📅 Published: 2026-06-30 📰 https://huggingface.co/blog/ibm-research/scarfbench 📌 Key takeaways: • ScarfBench evaluates AI agents on cross-framework Java migrations (Spring, Jakarta EE, Quarkus) — a fundamentally harder problem than bug fixing or code generation • Claude Code reported successful builds for 29 out of 30 applications, but independent verification showed many failures — agents are systematically overconfident about migration completion • Configuration dominates migration effort, not code translation; agents spend most of their time repeatedly returning to configuration artifacts resolving framework differences • Relevant to any team evaluating AI-assisted modernization: build success alone significantly overestimates migration quality; environment and tooling issues are the hidden blockers

• Can Ambient AI Make Classrooms Smarter? — Ambient AI that "fades into the environment" rather than sitting in a visible chat interface is emerging in education, with early pilots in engagement detection, adaptive pacing, and real-time instructional adjustment. 🔗 Graph: Higher Ed AI, AI Adoption 📅 Published: 2026-07-01 📰 https://edtechmagazine.com/higher/article/2026/07/can-ambient-ai-make-classrooms-smarter 📌 Key takeaways: • Ambient AI uses environmental sensors, device interaction data, and learning platform signals to detect engagement patterns — without requiring direct user interaction • Carnegie Mellon and Digital Promise have published research on ambient classroom sensing using voice and face recognition; Chinese universities are designing smart classrooms around the concept • Expert warns the technology can "look more precise than it actually is" — inferring attention or emotion from video/audio is not neutral and can embed bias • Privacy concerns are substantial: "Schools should be extremely skeptical of any implementation that feels like surveillance dressed up as personalization." Broader deployment expected in 2-5 years

• A Historian and Accidental Technologist's View of Higher Education's AI Moment — WCET's executive director puts the current AI wave in historical context beside the internet and MOOCs, arguing higher education must focus on AI literacy, critical thinking, and keeping humans in the loop. 🔗 Graph: Higher Ed AI, AI Governance, AI Adoption 📅 Published: 2026-07-02 📰 https://wcet.wiche.edu/frontiers/2026/07/02/historian-and-accidental-technologists-view-of-higher-education-ai-moment/ 📌 Key takeaways: • Van Davis traces higher ed's relationship with transformative technology from PLATO (1960) through the World Wide Web to MOOCs — noting each was predicted to radically reshape the model, but none fully did • Gallup (Oct 2025) found 79% of 18-28 year olds believe AI is making people lazier, and 62% believe it makes people less smart — even as those same students actively use AI • Argues the right question isn't whether AI will transform higher ed, but how to teach critical thinking, creative decision-making, ethical use, and when not to use AI • WCET is focusing on AI literacies, institutional policies, and emerging promising practices — the same challenge space Brett navigates with TritonAI governance

💡 Signal: Enterprise AI infrastructure is consolidating around private cloud for production inference — directly reinforcing TritonAI's on-prem architecture at SDSC. Meanwhile, research on coding agents shows they remain overconfident and struggle with real-world configuration complexity, a reminder that agentic AI promises still require human validation. In higher ed, the conversation is shifting from "should we use AI" to "how do we teach responsible use" — the exact governance frontier Brett is navigating.