Vendor and platform documentation on building and governing AI agents.
17 resources
AWS user guide for Amazon Bedrock Agents, covering action groups, knowledge bases, prompt overrides, guardrails, and multi-agent collaboration. Walks through IAM setup, tracing via CloudWatch, and invocation patterns for production agent deployments.
AWS product page for Bedrock AgentCore, presenting its runtime, gateway, memory, identity, observability, and browser-tool components. Positions AgentCore as framework-agnostic infrastructure for deploying agents built with any SDK at enterprise scale.
AWS News blog announcing AgentCore's general availability, detailing session isolation, identity-aware tool access, and observability primitives. Covers use cases, pricing model, and integrations with Bedrock Agents, Strands, LangGraph, and CrewAI.
Microsoft Learn documentation for Azure AI Foundry Agent Service, covering agent definition, threading, tool invocation, and content safety. Describes deployment options, Entra ID integration, and differences from the legacy Azure OpenAI Assistants API.
Microsoft's service card for Copilot Studio documenting system purpose, intended uses, capabilities, performance, limitations, and responsible-AI considerations. Serves as the transparency artefact for customers evaluating Copilot Studio for agent-building.
Microsoft's responsible AI overview and FAQs for Copilot Studio agent builders, covering grounding, content filters, prompt shields, user consent, data boundaries, and mitigation guidance for common failure modes in custom agents.
Microsoft's five-level agentic AI adoption maturity model spanning initial pilots through autonomous fleets. Details governance, security, data readiness, and change-management milestones at each level, with self-assessment and Copilot Studio-specific practices.
Microsoft's open-source AutoGen framework for multi-agent applications, offering a layered API (Core, AgentChat, Extensions) with async messaging, tool use, and group chat patterns. Includes AutoGen Studio, a no-code designer for multi-agent workflows.
Google Cloud codelab walking developers through building agents with Vertex AI Agent Builder, covering tool definition, reasoning loops, conversation management, and deployment. Uses Gemini models and integrates with Google Search and Dialogflow components.
Google's additional terms for the Gemini API governing developer use, covering acceptable use, user protections, restricted applications, prompt and output handling, and agentic use cases like browsing and transaction execution.
Anthropic's announcement of its updated Usage Policy with dedicated agentic AI provisions, covering authorisation, impersonation, scope boundaries, and user-consent expectations when Claude takes actions on a user's behalf through tools and computer use.
Anthropic support article translating the Usage Policy into operational guidance for agent builders, covering attribution, human oversight, prohibited autonomous actions, and how policy applies to long-running, tool-using deployments of Claude.
Anthropic's safety guide for Claude Cowork, covering scope of autonomy, credential handling, approval checkpoints, monitoring, and rollback when letting multiple Claude instances collaborate on tasks inside a user's environment.
OpenAI policy page for ChatGPT agent capabilities (browsing, terminal, computer use), detailing acceptable use, restricted actions (payments, sensitive accounts, account creation), and user-confirmation requirements for high-impact steps.
OpenAI developer documentation for Codex agent approval modes (auto, read-only, full auto) and security settings, covering sandboxing, network egress controls, secret handling, and configuring approval prompts for tool execution in coding workflows.
Salesforce developer guide explaining how Agentforce uses the Einstein Trust Layer - zero-retention prompts, dynamic grounding, toxicity detection, PII masking, and audit trails - to govern agents that act inside customer CRMs.
Databricks product page for Agent Bricks presenting it as a unified control plane for building and governing enterprise agents, with Unity Catalog permissions, AI Gateway policies, Mosaic AI evaluation, and observability tied to Lakehouse data.