How Stripe built “minions”: AI coding agents that ship 1,300 PRs per week. Stripe builds “minions” that convert Slack reactions into cloud-backed AI agents and ship roughly 1,300 reviewable PRs per week. This shows an agent-as-delivery-lane pattern where orchestration, review gates, and CI integration scale output — a blueprint for Principles 09 and 03 in practice.

Sycamore raises $65M seed to let enterprises build, deploy, and monitor AI agents. Sycamore raises $65M to build an enterprise agent operating system focused on building, deploying, and monitoring AI agents. Outcome engineers get a vendor-grade orchestration and observability stack to borrow from, shortening the work needed to standardize deployment, identity, and safety primitives (Principles 09 and 15).

OpenAI introduces a Codex plugin for Claude Code, letting users invoke Codex from inside Claude Code to review code or delegate tasks. OpenAI launches a Codex plugin that lets Claude Code call Codex directly for code review and delegated coding tasks. Multi-model, in-place delegation highlights practical model-composition patterns for building agent pipelines and handing off specialized skills (Principle 11).

Coasts — Containerized Hosts for Agents. Coasts runs isolated, containerized development hosts locally, booting reproducible workspaces for agent-driven workflows without any hosted service. That gives outcome engineers a repeatable local sandbox for debugging, security testing, and CI-like reproducibility when agents interact with real systems (Principles 07 and 06).

RSAC 2026 shipped five agent identity frameworks and left three critical gaps open. RSAC vendors ship multiple agent identity frameworks but fail to provide action-level audit trails of agents’ behavior, leaving endpoint observability gaps. Outcome engineers must treat identity as only the first step — instrument action auditing, tamper-resistant logs, and end-to-end observability before agents run at scale (Principles 13 and 15).