## Weekly signal

This was a practical week for AI agents in education and learning. The signal was not “one new tutoring app wins.” It was that institutions, course providers, and AI platforms are turning agentic AI into governed workflows, builder curricula, and assessment redesign. Coverage is through live search on May 11, 2026; no useful May 12-specific items were visible yet.

## What changed

1. OpenAI pushed agent-adjacent workflow tools deeper into education workspaces. ChatGPT for Excel and Google Sheets became available globally for Enterprise, Edu, and K-12 workspaces, with a free preview through June 2 and admin controls including RBAC, residency where available, EKM, and Compliance API coverage. OpenAI also added admin-console views for Analytics and Agents, giving owners visibility into usage, connected apps, schedules, memory files, and agent runs. For schools, the immediate use cases are budget models, enrollment trackers, advising spreadsheets, rubric analysis, and cleanup of messy operational data.

2. Agent-building education became more product-like. DeepLearning.AI released a short course on “Build Interactive Agents with Generative UI,” teaching developers to connect LangChain and Google ADK agents to React frontends via CopilotKit and AG-UI, so agents can render charts, forms, cards, and shared canvas interfaces rather than only text. That matters because learning agents often need inspectable, interactive workspaces, not black-box chat.

3. Universities are moving agentic AI from topic to curriculum. In India, IIIT Hyderabad opened applications for a 12-week online certificate on engineering agentic AI systems, covering design, implementation, testing, deployment, monitoring, AgentOps, MCP/A2A-style tool use, and human-in-the-loop design. In the U.S., ASU’s “The Agentic Self” course wrapped its first semester, with 75 learners building personal agent projects for small businesses, veterans’ benefits, and African-language learning support.

4. Assessment is becoming the pressure point. Oxford Brookes University in the UK hosted an internal session on AI-assisted coding, assessment design, and the agentic AI era. The takeaway was clear: assessments need to move toward problem framing, critical thinking, reflection, creativity, and applied work, because agentic systems can increasingly execute multi-step coding and knowledge tasks.

5. Security guidance is now part of education-agent adoption. Five Eyes cyber agencies published guidance urging careful adoption of agentic AI services, recommending low-risk tasks, least privilege, incremental deployment, continuous evaluation, and no broad access to sensitive systems. This is directly relevant to schools piloting advising, student-success, or LMS-integrated agents.

## What to do with it

Education teams should separate three tracks: teach agents, use agents, and govern agents. For teaching, update curricula around tool use, memory, evals, UI, deployment, and AgentOps. For institutional pilots, start with low-risk spreadsheet, scheduling, content-prep, and advising-support workflows. For governance, require an agent inventory, owner, data boundary, permission model, logging plan, rollback path, and human approval step before any student-impacting action.