From model to agent: Equipping the Responses API with a computer environment — OpenAI equips the Responses API with a sandboxed computer environment and shell tool so models can execute workflows and run commands in a controlled environment. That gives outcome engineers a production-ready runtime for agentic workflows with safer side-effect control, reproducible execution, and easier integration into orchestration layers (Principles 07 & 09).

Slashing agent token costs by 98% with RFC 9457-compliant error responses — Cloudflare now serves RFC 9457-compliant machine-readable error responses that cut agent token usage by over 98% and provide actionable retry guidance. This materially lowers operational cost for chatty agent loops, improves error-driven automation, and simplifies deterministic retry logic for orchestrators and state machines (Principles 06 & 11).

Manufact raises $6.3M as MCP becomes the ‘USB-C for AI’ powering ChatGPT and Claude apps — Manufact is building an open Model Context Protocol to plug AI agents into apps and raised $6.3M to accelerate adoption. A stable protocol for context and connectivity reduces bespoke integration work, making it easier to compose agents into product flows and transfer context across services for interoperable agent ecosystems (Principles 06 & 11).

Perplexity takes its ‘Computer’ AI agent into the enterprise, taking aim at Microsoft and Salesforce — Perplexity brings its multi-model “Computer” agent to enterprises, routing tasks across twenty models and isolating sessions with Firecracker microVMs. That demonstrates an enterprise blueprint for scaling agentic orchestration with model routing, tenant isolation, and hybrid model stacks — patterns outcome engineers can adopt for resilient, auditable systems (Principle 09).

Databricks launches data engineering copilot and acquires agent evaluation startup — Databricks unveils Genie Code, a data-engineering copilot, and acquires Quotient AI to evaluate and diagnose agent behavior. Outcome engineers gain both agent-assisted data pipelines to close deployment gaps and evaluation tooling to detect drift, failure modes, and misaligned behavior before it reaches production (Principles 03 & 16).