# Agent-executable business infrastructure Agent-executable office work, backed by business applications, tenant identity, saved state and explicit approvals. 100% agent-executable within the published scope means your connected agent can perform every supported execution step through tools, without manually operating the application screens. You provide the decisions and any required approvals; administrators connect your identity, models and external services first. ## Supported work today ### Create editable office files Create DOCX reports, XLSX workbooks with supported formulas, and PPTX presentations. Combine supported drafting steps into a saved multi-step plan. - Product: Workflow - Saved output: Private editable files, validation results, saved source revisions and authenticated download links. - Decisions and approvals: Review findings and the exact saved file separately. File approval does not send or publish it. - Setup and access: Authorized Workflow access and a configured model route for AI generation. Explicit-specification file creation does not require a model call. - Example tools: `workflow_plan_capabilities`, `workflow_create`, `workflow_prepare_office_artifact`, `workflow_inspect_office_artifact`, `workflow_approve_office_artifact` ### Review and revise documents Review supplied text, preserve the original, ask for missing information and prepare a revised document. Compare versions and retrieve the approved copy. - Product: Workflow - Saved output: Saved review findings, document versions, comparisons and approval receipts. - Decisions and approvals: Findings and revised document content have separate exact-version approvals. - Setup and access: Authorized Workflow access; configured tenant AI for model-generated reviews and revisions. - Example tools: `workflow_create`, `workflow_suggest_document`, `workflow_compare_document`, `workflow_approve_document`, `workflow_read_document` ### Turn requests into reviewed next steps Submit an available intake form, analyze the request, clarify missing details and prepare a Support request or restricted Passage project draft. - Product: Workflow - Saved output: Private intake receipts, saved analysis and an exact destination-and-fields preview. - Decisions and approvals: Submission-triggered analysis requires consent. Creating a destination record requires separate approval of the saved proposal and its audience. - Setup and access: An enabled intake form, eligible submitter/reviewer accounts and configured model routes. - Example tools: `workflow_available_intake_forms`, `workflow_submit_intake`, `workflow_request_information`, `workflow_prepare_action`, `workflow_approve_action` ### Plan and coordinate projects Create project drafts, propose plans, update saved tasks, move Kanban stages and collaborate on project documents and comments. - Product: Passage - Saved output: Project records, tasks, Gantt/Kanban reports, documents and revision history. - Decisions and approvals: Plan approval records the decision. It does not perform the real-world tasks listed in the plan. - Setup and access: Approved Passage access and project permissions; provider consent for supported file imports. - Example tools: `odyssey_create_project`, `odyssey_update_project`, `odyssey_move_task`, `odyssey_save_document`, `odyssey_project_report` ### Build structured learning Develop learning paths, course outlines and course content from a role or learning need, using the resources you are authorized to provide. - Product: Wayfinder - Saved output: Editable paths and courses with saved progress through the required review stages. - Decisions and approvals: Learning path, outline, content and publication decisions retain their required review gates. - Setup and access: Learning-authoring access, tenant quotas and an available model route for platform generation. - Example tools: `create_learning_path`, `get_course_outlines`, `get_course_pending_reviews` ### Move selected conversations forward Read selected Slack threads, prepare source-linked project suggestions, apply approved changes and send an exact approved reply to the original thread. - Product: Ahoy + Passage - Saved output: Saved thread reviews, project update receipts and tracked reply-delivery status. - Decisions and approvals: Sharing project updates and sending a reply require separate explicit authorization for the exact content, audience and destination. - Setup and access: The user’s connected Slack account, source-thread access and the required Ahoy/Passage tools. - Example tools: `ahoy_import_thread`, `ahoy_get_review`, `ahoy_apply_suggestions`, `ahoy_send_reply` ### Find the right people Search your tenant’s employee mirror by name, work email, title or department, and check directory synchronization status. - Product: Tenant identity - Saved output: Tenant-scoped work profiles and synchronization status, with no credentials returned. - Decisions and approvals: Agent directory tools are read-only. Employee lifecycle and group roles follow the configured source directory. - Setup and access: An administrator connects Google Workspace or Keycloak to the tenant’s tested company sign-in and approves the migration preview. - Example tools: `directory_search_employees`, `directory_get_status` ### Inspect observed agent activity Read collected execution traces, audit events and model-usage summaries for the activity your organization can access. - Product: Safe Harbor - Saved output: Recorded execution evidence, audit references and usage estimates. - Decisions and approvals: Read access follows tenant administration permissions. Coverage is limited to instrumented activity and submitted events. - Setup and access: Authorized Safe Harbor access and the relevant event/usage sources. - Example tools: `safe_harbor_list_traces`, `safe_harbor_audit_log`, `safe_harbor_cost_summary` ## Execution, approval and setup Your agent executes the supported operations above using the tools granted to its authenticated account. Required decisions and approvals can be supplied through the agent conversation where the approval tools are available. Service-agent draft enrollment does not grant every product or approval tool. Discover current access with authenticated MCP tools/list. An administrator establishes company sign-in, delegated directory access, model routes and provider consent. Google Workspace and tenant Keycloak lifecycle support is available after each tenant completes setup and validates its own connection. Directory provisioning and deactivation follow that configured authority; the agent directory tools provide lookup and status only. Multi-step drafting plans are bounded and use supported drafting/file capabilities. They do not turn free-text instructions into arbitrary tool execution. General-purpose mailbox/calendar automation, payroll, purchasing and unrestricted CRM/ERP operations are outside this scope. Safe Harbor observes instrumented activity and submitted events. ## Try it from the prompt ### One brief, three editable files Using the figures I provide, prepare a quarterly report, an editable workbook and a presentation. Show the saved drafts and validation before I approve the exact files. ### A policy review with a paper trail Review this policy, preserve the original and propose a revision. Compare the changes and ask for anything missing. Stop before approval. ### A request with a clear destination Review this request and propose a Support request or restricted project draft. Show the destination, fields and audience before creating the record. ### Colleagues without another search screen Find the onboarding contacts in my organization and show their work profiles. Check directory status without changing any accounts. [Connect your agent](https://robbclan.com/connect-your-agent) · [Discover current tools](https://robbclan.com/agent-docs) · [Task recipes](https://robbclan.com/agent-recipes) Examples are illustrative. Account permissions, tenant quotas, source access and configured model availability apply.