Agent Ops: orchestration, sandboxes, routing, and safe calls

Atlassian evolves Jira into an orchestration hub for developers and AI agents. Atlassian turns Jira into an orchestration hub, adding Planner and a Coding Agent to route work to developer and third‑party AI agents. This pushes issue trackers into the agent control plane and forces outcome engineers to embed routing, provenance, and delivery workflows where teams already plan work (Principle 09).

Perplexity launches SPACE sandbox to make its AI agents secure and powerful. Perplexity ships SPACE, a secure sandbox that lets Computer agents run at full capability while isolating risk. Treat this as a pattern for hosting risky tools and minimizing blast radius in production agent fleets (Principles 07, 14).

Model Routing Is Simple. Until It Isn’t.. IBM Research shows model routing demands multi-dimensional optimization—caching, infra constraints, governance, and cost overturn naive selection heuristics. Outcome engineers must design routing policies that balance latency, cache hit rates, trust boundaries, and fiscal controls, not just pick the highest-scoring model (Principles 06, 09, 11).

What building Shippy taught us about building agents. Shippy documents a production maritime agent built with versioned skills, deterministic tools, sandboxed hosting, and live-data validation. Use these artifact-level practices to make agent behavior auditable, testable, and safely deployable in domain-specific pipelines (Principles 07, 16).

Uncertainty Quantification for LLM Function-Calling. Apple presents methods to quantify function-call confidence and gate execution when models are uncertain. Apply these techniques to prevent irreversible actions, set execution thresholds, and instrument your gate for human oversight and audit (Principles 14, 15, 16).