- IBM and Microsoft announce a new AI agent platform designed for industry-specific transformation — IBM and Microsoft launched Imagine Enterprise, an agentic platform aimed at helping organizations build, run and govern AI agents tied to industry-specific processes. IBM Consulting contributes more than 6,000 prebuilt agents and assets built on its Consulting Advantage platform, while Microsoft supplies the foundation — Foundry, Copilot Studio, Fabric, Entra, Purview, Defender and Azure — plus forward-deployed engineers from both companies to move customers from experimentation into production. The joint announcement emphasizes governance: controls, observability and AI-compliance support built into the platform rather than added later, with regulated industries as the target. SEI is named as an early enterprise evaluating the platform.
📌 Key takeaways:
- Agent governance is being productized: identity, observability and compliance controls now arrive bundled with the agent runtime, so campus IT leaders should test vendor claims against the controls they already run — logging, spend tracking, data-boundary enforcement — instead of assuming a platform's dashboard satisfies institutional policy.
- Prebuilt agent libraries shift the build-versus-buy line for commodity processes, but the differentiator remains institutional data and process fit; the realistic campus play is adapting a small number of well-understood workflows rather than importing an industry catalog wholesale.
- Forward-deployed engineering is a services-heavy delivery model, so total cost includes implementation and integration labor — budget for that before committing, and expect the same model from other major vendors shortly.
🏛️ UCSD angle: UCSD keeps model and agent choice behind an enterprise LiteLLM gateway that handles routing, failover, spend tracking and audit logging, with use cases clearing a campus Data Governance Council — the same "govern the agent, not just the model" structure vendors are now packaging into products.
- Agent Architecture Cost-Success Trade-offs: A Controlled Evaluation across Enterprise-Style SOPs — A 14-author study held the model, tasks and tools fixed and varied only the agent architecture, comparing a function-calling single agent, a skill-augmented agent, centralized multi-agent orchestration and a triage swarm (eight arms) across 13 enterprise-style standard operating procedures. Three findings matter for buyers: there is no universally best architecture and performance is SOP-dependent; every coordinated arm consumed 1.4–4.7x the tokens and 2.6–5.8x the latency of a single agent for gains that were usually small and rarely statistically significant; and better skill definitions lifted task success from 67.4% to 73.1% without changing architecture at all.
📌 Key takeaways:
- The single agent is the rational default: research universities and colleges planning multi-agent pilots should require evidence of a task-success gain before paying the two-to-five-fold token and latency premium that orchestration adds.
- Improving instructions and skill definitions is the cheapest available lift — a measurable accuracy gain for authoring effort, with no architectural rework or additional inference spend.
- Instrument cost per successful task, not cost per call; a cheap agent that fails becomes expensive once retries and human cleanup are counted, and that ratio is the number governance committees should see.
🏛️ UCSD angle: UCSD's agent work is gated by an evaluation function that runs shadow-mode and cohort-split trials with per-user token logs and model routing pulled from the campus gateway before any agent scales campuswide — the measurement discipline this paper argues for.
- Phishing Threats Reached 7.3 Billion in Q3 2026 — Microsoft Threat Intelligence's quarterly analysis counted roughly 7.3 billion email phishing threats in Q3 2026, with detections reaching nearly 3 billion in September alone and September activity up 41% from August. Attackers leaned on business-service impersonation — about 129 million messages spoofing tracked business-service domains, with DocuSign the most spoofed brand in September at 29% — and on Teams, where Microsoft recorded roughly 141,000 malicious call attempts and saw attacker display names shift from fake IT support to ordinary personal names (46% of a sample in July, 88% by September). Critically, some malicious messages were delivered despite failing authentication checks and being classified as threats: tenant- or user-level security overrides applied to 63% of those messages with a recorded outcome.
📌 Key takeaways:
- Campus security teams should audit tenant-level allow lists, transport exceptions and authentication overrides before buying anything new — the report's most actionable finding is that well-intentioned exceptions, not detection failures, are letting impersonation mail through.
- Collaboration platforms are now a first-class phishing surface: voice and chat impersonation using ordinary human names defeats training built around "fake IT support" scripts, so help-desk and faculty awareness content needs updating.
- Credential theft remained the objective in 94–96% of payload-based campaigns, which keeps phishing-resistant authentication and conditional access the highest-leverage controls a university can fund.
🏛️ UCSD angle: UCSD runs a Phishing Analyzer agent inside its TritonAI portfolio to triage suspicious messages, putting AI on the defensive side of the same capability curve this report describes.
- Matthew Pittinsky Returns to the Helm at Blackboard — Blackboard co-founder Matthew Pittinsky has returned as CEO, completing a transition that began when Anthology emerged from Chapter 11 bankruptcy and rebranded as Blackboard to refocus on its core teaching and learning business. Pittinsky put AI at the center of his stated vision, arguing that "the learning management system is still largely a course management system" and that AI offers a chance to deliver what a true LMS should. He is joined by a new CTO, Jason Simpson, hired in September with a mandate centered on platform performance, reliability and security, and a new chief legal officer.
📌 Key takeaways:
- A vendor emerging from bankruptcy with a new CEO is a contract-risk moment: institutions running the platform should re-read termination, escrow, data-extraction and roadmap-commitment clauses while leadership is still in listening mode.
- Expect AI features bundled into the LMS core and priced accordingly; ask for evaluation evidence and student-data terms before piloting, because an AI feature inside the LMS inherits whatever data governance the LMS contract allows.
- The CTO hire signals that reliability and security are acknowledged problems, not marketing themes — a reasonable first ask for any campus is a written remediation plan with dates.
🏛️ UCSD angle: UCSD keeps its AI layer LMS-adjacent rather than LMS-embedded — course-content retrieval runs through Onyx and the campus gateway, with Canvas API integration work in progress — so the model layer can change without renegotiating the learning platform.
- Data centers need power – lower income, rural residents pay the price — Cornell research published in Nature Energy analyzed energy demand and infrastructure projects across four major U.S. power grids from 2020 to 2025 and found that data centers paid only 38% of the infrastructure costs linked to their growth in 2025 — $3.12 billion of an $8.25 billion bill — leaving the remaining 62% to other ratepayers. Lower-income and rural households were hit hardest, with the lowest-income households seeing extra energy costs reaching up to 28% of their incomes. Large data centers can consume as much power as medium-sized cities, and the finding lands as campuses pursue their own AI compute build-outs.
📌 Key takeaways:
- Research universities negotiating large compute or colocation deals should ask who pays for grid upgrades before signing: cost allocation is becoming a public-policy question, and institutions can inherit both the bill and the reputational exposure.
- Expect rate cases, state legislation and utility commissions to take up data-center cost shifting — campus government-relations and facilities teams should be briefed on the institution's own AI energy footprint now, not after a filing names it.
- Publishing an honest energy and cost accounting alongside AI infrastructure growth is cheaper than defending its absence; the equity framing in this study will be used by critics of campus compute expansion.
🏛️ UCSD angle: UCSD expanded its shared on-campus GPU platform this month to absorb student AI demand that had been spilling into commercial cloud, keeping that growth inside existing campus facilities and cost structures rather than adding third-party capacity.
- Classroom Technology Standardization Transforms the Faculty Experience — EdTech Magazine profiles the University of Rhode Island's multi-year move away from audiovisual components sourced from many manufacturers toward a single AV-over-IP platform, now in roughly half of its classrooms. Standardization lets the IT team see system status and power-cycle equipment remotely, cut the number of spare parts on hand, and free up repair budget; when the first standardized rooms opened in fall 2024, the feared flood of support calls never arrived. That operational relief has turned into demand — the most common instructor comment is a request for the same setup in their other rooms.
📌 Key takeaways:
- Treat the classroom fleet as a platform: a published reference design, a single remote-management plane and a standing refresh cycle cut support hours and parts inventory more reliably than per-room upgrades.
- Write remote telemetry and remote power cycling into AV procurement requirements — the operational payoff in this case came from being able to diagnose and fix rooms without a truck roll.
- "When is this coming to my room?" is a better adoption metric than satisfaction scores, and it gives IT leaders a defensible way to sequence the remaining classrooms.