- Uber puts all 30,000 employees on ChatGPT Work — TechCrunch reports that Uber is scaling OpenAI's enterprise agent platform ChatGPT Work from a 1,000-person pilot to its entire global workforce of roughly 30,000, marking the company's first company-wide deployment of an agentic platform for knowledge work. Notably, about 80% of the pilot's early adopters were non-engineers — the tool was used to analyze operational data, localize training materials across markets, and build personalized dashboards. Uber is pairing the rollout with an internal library of AI skills, approved integrations, and a plan to measure how workflows actually change before deciding what comes next, rather than treating licenses as the deliverable.
📌 Key takeaways:
- For institutional leaders weighing campus-wide AI deployments, Uber's sequencing is the real lesson: pilot broadly across job families, embed technical experts with functional teams, and scale only after usage patterns are understood — not license first and ask questions later.
- The 80% non-engineer adoption figure undercuts the assumption that agentic tools are developer territory; business and administrative workflows are where these platforms are getting traction first.
- Uber's explicit "we'll evaluate outcomes and use what we learn" stance is a template for institutions under pressure to justify AI spend: define the measurement plan before the rollout, not after.
🏛️ UCSD angle: UCSD's campus AI platform has followed the same pattern at university scale — broad access for staff and faculty beyond technical users, with adoption tracking informing expansion decisions — and Uber's AI-pods model mirrors the campus's forward-deployed engineering approach for turning platform access into real workflows.
- NOT-OD-27-011: USG Policy for the use of Super Intelligence — NIH released an official notice confirming that the federal government is transitioning from "Artificial Intelligence" and "AI" to "Super Intelligence" and "SI" in official documents and communications — including grant notices, websites, and reports — pursuant to Section 2 of Executive Order 14434, signed September 29. The notice assigns the new label the existing statutory definition of artificial intelligence in 15 U.S.C. 9401(3), meaning nothing substantive changes about what the term covers; only the terminology does. NIH grants pages already carry transition banners, and the change does not require alterations to active awards, though new funding opportunities may adopt the new language.
📌 Key takeaways:
- Research universities should expect the terminology shift to propagate through federal grant solicitations, compliance documents, and agency reporting in the coming months — research administration and IT communications teams need a plan for when institutional language follows federal language and when it doesn't.
- Because the new term inherits the existing statutory definition, no policy, IRB, or procurement obligations change — leaders should resist the urge to re-open settled AI governance work over a rebrand, while updating external-facing documents where consistency with funders matters.
- Institutions with active federal awards should note the explicit carve-out: nothing in the order requires altering existing awards, so this is a communications burden, not a contractual one.
🏛️ UCSD angle: UCSD's AI governance structures — which deliberately sit separate from any single vendor or federal framing — are well positioned to absorb a terminology change like this without reopening policy; the practical impact lands in sponsored research communications, where the campus already runs dedicated compliance channels.
- LitPivot: Developing research ideas through dynamic contextualization within the literature landscape — The Allen Institute for AI (AI2) released LitPivot, a research ideation assistant that helps researchers refine early-stage research ideas by situating them in the live literature — generating literature-supported critiques and revisions as a proposal evolves rather than producing one-shot feedback. Unlike static literature-review tools, LitPivot continuously contextualizes the idea against the research landscape as it changes, positioning itself as a companion for the messy early phase of research design where most AI writing tools don't operate. The release continues AI2's run of openly licensed research tooling aimed squarely at academic workflows.
📌 Key takeaways:
- Research universities building campus AI support programs should watch the ideation-assistant category mature: proposal development and literature contextualization are becoming distinct AI-assisted workflows, separate from writing and analysis tools.
- AI2's pattern of open research tooling matters for institutions with data-residency or cost constraints — openly licensed tools can be evaluated and deployed inside institutional AI gateways where commercial equivalents can't.
- Faculty development programs should note the workflow shift: these tools intervene at the idea-formation stage, which means AI literacy training needs to cover research design, not just prompt-and-draft.
🏛️ UCSD angle: Openly licensed tools like this are natural candidates for the campus's model-agnostic AI gateway, which was built precisely to evaluate and serve multiple providers and open models under institutional governance rather than betting on a single vendor.
- Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II) — OpenAI's infrastructure team published the second installment of its storage-scaling series, detailing how its Habitat storage layer was reworked with reliable multi-tenancy, global caching, and Azure Cosmos DB to deliver 12x faster reads for ChatGPT at planetary scale. The post walks through the operational consequences of serving a billion-plus users — data partitioning, tenant isolation, and the read-heavy access patterns that dominate consumer AI workloads — and how the architecture evolved from the simpler design described in Part I. It is one of the more detailed public engineering accounts of what consumer-AI-scale storage actually requires.
📌 Key takeaways:
- Campus infrastructure teams planning AI platform storage should steal the architectural lessons even at smaller scale: read-heavy caching layers and multi-tenant isolation are exactly the access patterns institutional AI platforms hit as adoption broadens.
- The post is a useful calibration point for capacity planning — consumer AI workloads generate radically skewed read patterns that traditional storage sizing models underestimate.
- Multi-tenancy is treated as a first-class design constraint, not an afterthought — the same discipline institutions need when one platform serves faculty, staff, students, and researchers with different data boundaries.
- How AT&T is rethinking in-house legal work with OpenAI — OpenAI profiles AT&T's LegalEdge team, which is using ChatGPT to rebuild how legal knowledge is accessed inside the company — rethinking intake, surfacing institutional legal knowledge, and scaling legal judgment across a large distributed legal function. The case study focuses less on document review automation and more on knowledge access: the bottleneck the team attacked was lawyers' ability to find and reuse the organization's accumulated legal expertise. AT&T describes the work as augmenting judgment rather than replacing it, with intake triage as the first workflow to change.
📌 Key takeaways:
- The knowledge-access framing transfers directly to higher ed: university legal, procurement, and policy offices hold decades of institutional precedent that staff routinely fail to find — AI-assisted retrieval over that corpus is a high-value, low-controversy starting use case.
- Intake triage is emerging as the canonical first workflow for professional-service AI adoption across sectors; institutions designing their own AI programs should look at their intake processes as candidate quick wins.
- AT&T's pairing of technology rollout with an explicit "scaling judgment, not replacing it" narrative is a communications model institutional leaders should borrow for anxiety-heavy deployments.
🏛️ UCSD angle: This is the same play UCSD has already run with its automated contract review workflow — AI applied to institutional document processes to cut turnaround time dramatically — and the AT&T intake-triage pattern is a proven template for the next business-office process candidate.