AI.info
AI agents
Action under control: goals, tools, permissions, stop conditions, and who answers for what the system did.
- Agentic Systems as Controlled Action
- Workflow, Assistant, or Agent?
- Goals, Task Contracts, and Success Criteria
- Autonomy, Authority, and Delegation
- Environment, Observation, State, and Action
- Agent Loops, Harnesses, and Runtimes
- Deterministic Control and Model Judgment
- Architecture Decision Map for Agentic Systems
- Function Calling and Typed Action Contracts
- Tool Schema Design and Agent Affordances
- Tool Discovery, Selection, and Routing
- Tool Results, Provenance, and State Updates
- Timeouts, Retries, and Idempotent Tool Use
- Transactions, Side Effects, and Compensation
- Credentials, Scopes, and Least Privilege
- Search and Retrieval Tools for Agents
- Browser Agents and Web Navigation
- Code Execution and Sandboxed Compute
- Computer Use and GUI Control
- Model Context Protocol: Tools, Resources, and Prompts
- Agent2Agent Protocol and Remote Agent Collaboration
- Tool Contract Testing and Tool-Use Evaluation
- ReAct and the Action–Observation Loop
- Task Decomposition and Dependency Graphs
- Plan-and-Execute Versus Stepwise Control
- Replanning, Branching, and Search Budgets
- Verifiers, Tests, Critics, and Outcome Checks
- Reflection and Self-Correction Without Magical Thinking
- State Machines, Policies, and Explicit Orchestration
- Parallelism, Concurrency, and Synchronization
- Clarification, Uncertainty, and Abstention
- Long-Horizon Work, Checkpoints, and Resumability
- Termination, Loop Detection, and Safe Stopping
- Context Engineering and Attention Budgets
- Working Memory, Scratch State, and Task Ledgers
- Episodic Memory and Event Histories
- Semantic Memory and Knowledge Stores
- Memory Write Policies, Retrieval, and Forgetting
- Summarization, Compaction, and Context Drift
- Durable State, Provenance, Replay, and Recovery
- Identity, Privacy, Retention, and Deletion
- When Multi-Agent Systems Help—or Hurt
- Supervisor–Worker Orchestration
- Handoffs, Specialists, and Capability Routing
- Parallel Research and Map–Reduce Patterns
- Critique, Debate, and Result Aggregation
- Shared State, Messaging, and Coordination
- Deadlocks, Cascading Errors, and Conflict Resolution
- Human–Agent Teams, Escalation, and Mixed Initiative
- Research Agents and Evidence Synthesis
- Coding Agents and Repository Workflows
- Transactional and Customer-Service Agents
- Agent Evaluation Starts With Environment State
- Trajectory, Tool, and Policy Evaluation
- Reliability, Simulation, and Benchmark Design
- Observability, Tracing, and Systematic Agent Debugging
- Prompt Injection, Secrets, and the Confused Deputy
- Capability Control, Approvals, Sandboxing, and Rollback
- Cost, Latency, Deployment, Monitoring, and Incidents
- AI Agents Capstone: Design, Evaluate, and Govern an Agentic System