OpenTelemetry Export for GitHub Copilot App Agent Sessions
Sep 25, 2026·12 min·1 plays
The question this episode answers
Can an organization enable centralized OpenTelemetry for GitHub Copilot App Agent sessions and still investigate agent failures without unnecessarily exporting or retaining sensitive prompt, response, and tool data?
Organizations can now trace Copilot model requests and tool activity centrally, so review telemetry retention, endpoint access, content-capture settings, and whether traces are sufficient to investigate agent failures without exposing prompts or responses. ([github.blog](https://github.blog/changelog/2026-09-22-opentelemetry-in-the-github-copilot-app/))
#ai-builders#hook-family:risk
Researched & cited
Every claim is researched from multiple sources and cross-checked before narration. Daycast is made with AI — researched, written, and narrated — and we say so in every episode. Sources below.
Engineering organizations should review merge protections, review budgets, and repository policies before GitHub changes the default effort level to Balanced on September 28, 2026, while Copilot approvals can already count toward required approvals when enabled. ([github.blog](https://github.blog/changelog/2026-09-01-copilot-code-review-can-now-approve-pull-requests/?utm_source=openai))
OpenAI now manages session orchestration, context compaction, recovery, and sandbox integration, so teams should review whether the new convenience layer creates unacceptable control, portability, or observability boundaries. ([openai.com](https://openai.com/index/introducing-the-agents-api/?utm_source=openai))
Anthropic’s expanded review found that sealed-evaluation assumptions, third-party harness configuration, and incomplete transcript search can combine into real-world access, so AI teams should require network enforcement, live monitoring, and independently verifiable evaluation boundaries. ([anthropic.com](https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents?utm_source=openai))
Teams using `antigravity-preview-05-2026` with local tools must review parsers, tool wrappers, and file-edit assumptions before the October 5, 2026 shutdown of the old agent.
