- DevDay 2026 Recap — OpenAI's developer conference delivered more than 20 announcements, headlined by dots: always-on agents running on GPT-6 Astra with their own cloud computer, plugin access to 4,000+ apps, and presence inside ChatGPT, Slack, and Teams. Enterprise, Edu, and Healthcare workspaces get dots in beta, off by default until an administrator enables them. The rest of the stack matters for campus builders too: an Agents API with computer use and context compaction hosted by OpenAI, Codex Security Cloud for repository scanning, a Decisions API for routing and classification, and Sign in with ChatGPT, which lets users spend their plan allowance across 16 partner tools.
📌 Key takeaways:
- Campus IT leaders should note that the always-on agent features arrive in Edu workspaces off by default, which puts the enable/disable decision squarely on institutional administrators rather than vendors.
- The hosted Agents API shifts multi-agent plumbing (scheduling, tool routing, context compaction) from something organizations build to something organizations rent, changing the build-vs-buy calculus for campus AI teams.
- Sign in with ChatGPT quietly moves token costs onto individual user plans, a procurement and licensing wrinkle worth watching as partner tools multiply.
🏛️ UCSD angle: UCSD's model-agnostic LLM gateway approach means new model and agent capabilities like these can be evaluated and routed as they ship, rather than requiring a vendor lock-in decision.
- Introducing GPT-6.1 Sol — OpenAI's new mid-tier model delivers near-Astra intelligence for coding, computer use, and professional work at one-fifth of Astra's standard token prices: $2 per million input tokens, $10 per million output, and cached input at $0.10. OpenAI says Sol matches Astra on DeepSWE, a long-horizon software engineering benchmark, and outside tests put its cost per benchmark task at roughly a fifth of Astra's. The aggressive cached-input price is the sleeper item: agent workloads that resend long system prompts and tool lists every turn are where the bill actually drops.
📌 Key takeaways:
- The price-to-capability curve is falling fast enough that campus AI platforms can now route routine agentic and coding work to a near-frontier model at a fraction of flagship cost.
- Cached-input pricing at $0.10 per million tokens specifically rewards well-architected agent workloads, so institutions investing in prompt-caching-friendly designs will see the biggest savings.
- Budget planners should treat per-token list prices as a moving target and build recharge or cost-allocation models that flex with them.
- Higher Ed's AI Integration Overhyped, New Data Shows — For the first time, Instructure analyzed which external learning tools higher education users actually integrated into Canvas LMS, and the results undercut the AI adoption narrative: the most popular AI tool didn't even crack the top 100 integrated tools. The data suggests that headline claims about AI transforming teaching and learning are running well ahead of what measured usage inside the LMS shows. It's one of the first hard, platform-level looks at real adoption behavior rather than survey sentiment.
📌 Key takeaways:
- Technology leaders should insist on platform-level usage telemetry, not vendor projections or faculty survey sentiment, when deciding which AI investments deserve budget.
- The gap between perceived and measured AI adoption is a caution against procurement driven by hype cycles; pilot programs with defined usage metrics will age better than enterprise licenses bought on narrative.
🏛️ UCSD angle: UCSD's campus AI program has emphasized measurable adoption and utility over volume of deployments, an approach this data validates.
- College leaders are using AI. Students call it hypocrisy. — Dartmouth said it would investigate its provost over AI accusations, and similar controversies on other campuses have prompted frustration among students who face strict AI rules in the classroom. The piece captures a growing credibility problem: institutions publicly urge faculty and students to embrace or restrict AI, but leadership's own AI use is drawing scrutiny when policies feel inconsistently applied. The discordance lands hardest at institutions, like Dartmouth, that have made public pushes for AI in higher education.
📌 Key takeaways:
- AI policies that bind students but not leadership invite legitimacy challenges; institutions should apply the same disclosure and integrity standards across the org chart.
- Governance committees should clarify how AI-use rules apply to administrative and executive work product, not just coursework, before an incident forces the question.
- Digital accessibility deadline demands a focus on procurement — With roughly eight months until the ADA Title II digital accessibility compliance deadline, the piece argues that procurement is where institutions will win or lose: contracts, vendor accessibility documentation, and purchasing processes are the leverage points for actually achieving compliance rather than chasing retrofits. The timeline is short enough that institutions without a procurement-focused remediation plan are already behind.
📌 Key takeaways:
- Campus technology leaders should audit current procurement templates now for accessibility requirements, since new contracts signed before the deadline inherit whatever gaps exist.
- Institutions that centralize vendor accessibility documentation (VPATs, conformance claims) in the purchasing workflow will avoid a scramble to remediate systems bought without those terms.
- AI needs $6 trillion in annual revenue to justify data centers — Bain's annual global technology report argues the AI industry must generate $6 trillion in annual revenue by 2031 to justify the capital being deployed on data center construction worldwide, with existing consumer and enterprise AI services contributing only about $1.8 trillion of that. Bain projects $5 trillion to $6.5 trillion in data center spending by 2030, adding roughly 150 gigawatts of capacity, and notes that local opposition and power, equipment, and labor bottlenecks are already delaying billions in U.S. projects. The report frames the buildout as running well ahead of the demand curve, with sustainability hinging on revenue sources, from robotics to drug discovery, that barely exist yet.
📌 Key takeaways:
- Research universities planning AI infrastructure investments should assume sustained upward pressure on compute, power, and cooling costs, and stress-test cloud commitments against the possibility of an infrastructure correction.
- If the $4.2 trillion revenue gap forces a slowdown or repricing of capacity, institutions with flexible, multi-vendor compute strategies will be far better positioned than those locked into long-term single-provider commitments.