Build an agent? Sell an agent reports Microsoft opening its Marketplace to AI agents so developers can publish, monetize, and surface context-aware agents across Microsoft platforms. That matters because treating agents as first-class, distributable artifacts reshapes how you package skills, enforce governance, and operate delivery lanes — Principle 08/09.

Microsoft open sources ASSERT, an AI evaluation framework for enterprise agents shows ASSERT turning written requirements into executable tests for agent evaluation and regression pipelines. Outcome engineers can embed those tests into CI/CD to catch behavioral regressions and codify acceptance criteria for production agents — Principle 14/16.

TestSprite launches an open-source command-line tool to help AI agents check their own work announces a CLI that lets coding agents automatically verify and test their own code. Adding a lightweight self‑check loop reduces manual review and ‘botsitting,’ letting you close the feedback loop between agent outputs and test artifacts — Principle 14/16.

Diagrid brings cryptographic proof to AI agent and workflow execution covers Dapr 1.18’s verifiable execution features that cryptographically prove provenance, custody, and tamper-evident history for agent workflows. Verifiable execution gives you auditable evidence of what agents did and when, enabling stronger validation, forensics, and compliance controls — Principle 02/16.

Google DeepMind warns about risks as millions of AI agents begin interacting reports DeepMind funding $10M to study multi-agent risks as scaled agent interactions could enable coordinated harms. You must design orchestration with safe defaults, monitoring, and throttles because emergent multi-agent behaviors change threat models for deployed systems — Principle 09/14.