Atlassian embeds agents into Jira and embraces MCP for third-party integrations. Atlassian installs AI agents inside Jira and opens Rovo to third-party agents via the MCP protocol in an open beta. Outcome engineers should treat work systems as agent platforms—plan for embedded agents and standardized tool catalogs to enable cross-agent workflows (Principles 03, 09).

Perplexity launches Perplexity Computer, “a general-purpose digital worker” that can route work across 19 AI models. Perplexity ships a digital worker that routes tasks across multiple models as a single product for subscribers. That shifts the execution layer from single-model agents to multi-model orchestration—rethink routing logic, model-selection heuristics, and observability for outcome guarantees (Principles 09, 03).

Vast Data expands AI Operating System with global control plane, zero-trust agent framework and deeper NVIDIA integration. Vast adds a global control plane and a zero-trust agent framework to its AI OS for hybrid multicloud deployments. Treat this as a blueprint for production agent fleets: enforce zero-trust, centralize policy and telemetry, and design control-plane APIs to audit and gate agent behavior (Principles 09, 10).

Poisoning AI Training Data. Bruce Schneier shows a 20‑minute fake webpage can corrupt web-sourced training data and hijack chatbot outputs. Outcome engineers must assume training and retrieval surfaces are adversarial—build provenance, ingestion filters, and continuous validation to detect and quarantine poisoned signals (Principles 02, 14, 16).

Anthropic acquires Vercept to bring Vy’s natural-language desktop control to Claude. Anthropic integrates Vercept’s Vy to give Claude natural-language desktop automation and control. If your agents interact with user desktops or enterprise endpoints, treat desktop-level capabilities as a feature and a risk—design identity, least-privilege gating, and audit trails before enabling autonomous desktop actions (Principles 03, 09).