workflow orchestration, document processing, human-review steps and appropriate controls. Build reusable agent harnesses: practical runtime patterns around tools, state, context, approvals, traces and test fixtures that make agents consistent, inspectable and safe to use. Build grounded retrieval and context patterns using approved data products, search indexes, document repositories … users understand AI output, inspect supporting evidence, provide feedback and complete review or approval steps. Write maintainable Python and TypeScript code; use Docker, Git, automated testing and CI/CD to make development, testing and deployment repeatable. Apply agreed security, authentication, data-handling, deployment and release patterns, working with ...