PTIR — Daily Briefing — November 20, 2026
by
Executive Summary
College Board has published the first detailed preview of an AI-centered AP Computer Science Principles redesign for fall 2027. Docker is recasting its developer platform around isolated, governed AI-agent execution, while GitHub Actions has moved RHEL 9 and 10 larger-runner images into general availability.
Action Queue
1. Review the AP CSP redesign before the final framework arrives
The redesigned course makes responsible AI, supervised hands-on labs, verification of AI-generated code, attribution, privacy, and human decision-making part of the assessed curriculum. Its new AI Create Project asks students to build an AI-powered application for a real audience; a finalized course and exam description is due in spring 2027.
- Action: Read the framework overview and note the distinction between code created with AI and an application that actually uses AI functionality.
- Urgency: This week
- Importance: ★★★★★
- Verified active: November 20, 2026
- Deadline: No immediate deadline; redesigned course begins fall 2027
- Cost: Free
- Requirements: Web access; districts will eventually need an approved AI tool and individual student access
- Official source: College Board AP CSP redesign
2. Treat AI agents as actors, not ordinary workloads
Docker’s current security argument is that a coding agent can decide what to execute, access, and change, so ordinary container boundaries are not the whole control plane. Docker Sandboxes use a separate microVM and private Docker Engine for each agent session; governance adds centrally managed filesystem, network, MCP, and audit policies.
- Action: Read the architecture comparison and record which files, network destinations, secrets, and tools an agent actually needs before granting access.
- Urgency: This week
- Importance: ★★★★☆
- Verified active: November 20, 2026
- Deadline: None
- Cost: Local Docker Sandboxes are free; cloud sandbox compute starts at $0.07/hour and AI Governance is custom-priced
- Requirements: Supported macOS, Windows, or Linux environment; model/API access is separate
- Official sources: Docker Sandboxes versus containers and Docker AI Governance

3. Use RHEL-native GitHub runners when production parity justifies the cost
RHEL 9 and RHEL 10 images for GitHub-hosted larger runners are now generally available and fully supported. They allow workflows to test on the same operating-system family used in production, or to serve as a base for controlled custom images.
- Action: Save the announcement for projects that deploy to RHEL; keep standard runners when distribution parity does not materially improve the test.
- Urgency: No deadline
- Importance: ★★★☆☆
- Verified active: November 20, 2026
- Deadline: None
- Cost: Requires paid GitHub Team or Enterprise Cloud access and billable larger-runner usage
- Requirements: Organization access capable of creating Linux x64 larger runners
- Official source: GitHub Changelog
AI
The AP CSP preview offers a useful test for AI curriculum: students should be able to explain what the system does, evaluate generated code and output, choose an appropriate degree of automation, document outside help, and defend the application’s treatment of real users.
Linux & Self-Hosting
Docker’s sandbox-versus-container distinction is durable beyond one product. Containers package workloads; a tool-using agent also needs explicit limits on credentials, files, network access, external tools, and post-run evidence.
Teaching Corner
What Counts as an AI Application? gives students three short scenarios—AI-assisted coding, an app calling an AI service, and a student-built model—and asks them to classify each, identify evidence, and name correctness and safety tests.
GitHub Discoveries
RHEL larger-runner availability closes a gap for teams that need production-parity CI on Red Hat’s supported enterprise distributions. It is a targeted option rather than a default: the value depends on whether OS-specific behavior is genuinely part of the deployment risk.
PKb Candidates
Agent permission envelope: Record an agent’s allowed files, network destinations, secrets, tools, approval points, and audit evidence before execution.
AI application test: Separate who wrote the code from what the finished program does; document the model or AI service, user-facing decision, inputs, outputs, evaluation method, and safety boundary.
tags: artificial-intelligence - technology-education - programming - docker - github-actions - linux