Artificial Intelligence
Microsoft AI-103: Multi-Agent Orchestration
A support investigation needs three kinds of evidence: account activity, shipping records and a contract interpretation. Splitting the work may help if…
Read article05 / THE EDITORIAL LIBRARY
Article archive
New thinking and practical technical perspectives
Artificial Intelligence
Microsoft AI-103: Building Agents in Foundry
A claims assistant may read a policy, compare records and draft a decision, but it must not approve its own payment. The…
Read articleArtificial Intelligence
Microsoft AI-103: RAG and Vector Search
An internal policy assistant answers the same question correctly on Monday and incorrectly on Friday. The language model has not necessarily changed;…
Read articleArtificial Intelligence
Microsoft AI-103: Foundry Service Selection
A support team wants to answer employee policy questions, extract facts from uploaded documents and let a case worker approve certain actions.…
Read articleArtificial Intelligence
Anthropic CCA-F: Testing Claude Agents and Guardrails
An agent can produce the right answer for the wrong reasons. It might correctly flag an expense claim while reading another employee’s…
Read articleArtificial Intelligence
Anthropic CCA-F: Choosing the Right Claude Agent Architecture
Architecture discussions often begin with model size or the number of agents. Neither is the right starting point. Consider an access-review assistant…
Read articleArtificial Intelligence
Anthropic CCA-F: Context Management in Claude Agents
Long context does not guarantee accurate context. A Claude assistant may reopen a case with thousands of tokens of history, including superseded…
Read articleArtificial Intelligence
Anthropic CCA-F: Reliable Structured Outputs with Claude
A language-model response can look correct to a person while being unusable to software. An extraction service may return elegant prose instead…
Read articleArtificial Intelligence
Anthropic CCA-F: Claude Code Workflows That Hold Up
Claude Code can inspect a repository, propose edits, execute permitted commands and iterate when tests fail. That speed is valuable only if…
Read articleArtificial Intelligence
Anthropic CCA-F: Recovery from Agent Tool Failures
A missing tool response can hide two very different events: an operation that never started and an operation that succeeded before the…
Read articleArtificial Intelligence
Anthropic CCA-F: Designing Safer MCP Tools for Claude
Connecting Claude to an external system looks straightforward until the model can do more than read information. A customer-support assistant may need…
Read articleArtificial Intelligence
Anthropic CCA-F: Orchestrating Claude Subagents
Multi-agent systems create a new management problem as soon as they split the work. A coordinator can ask separate Claude workers to…
Read article