https://aiagentnook.com/ AI agents from first principles: what they are, the types, how to evaluate them, and the build flow for shipping one. en Wed, 12 Aug 2026 12:00:00 +0800 https://aiagentnook.com/what-is-an-ai-agent.html https://aiagentnook.com/what-is-an-ai-agent.html Wed, 12 Aug 2026 09:00:00 +0800 A definition narrow enough to be useful: an AI agent is a model using tools in a loop it controls itself. What that means, and why autonomy is a cost, not a feature. https://aiagentnook.com/types-of-ai-agents.html https://aiagentnook.com/types-of-ai-agents.html Wed, 12 Aug 2026 09:10:00 +0800 The textbook taxonomy of AI agents and the workflow-versus-agent split practitioners actually use, plus the environments that shape coding, browser, and voice agents. https://aiagentnook.com/ai-agent-evaluation.html https://aiagentnook.com/ai-agent-evaluation.html Wed, 12 Aug 2026 09:20:00 +0800 Task success rates, the public benchmarks (SWE-bench, GAIA, tau-bench, OSWorld), building your own eval suite, and what a success rate hides. https://aiagentnook.com/build-flow-and-sequence.html https://aiagentnook.com/build-flow-and-sequence.html Wed, 12 Aug 2026 09:30:00 +0800 The seven-step order of operations for shipping an AI agent — task, baseline, tools, loop, evals, guardrails, deploy — and why the sequence is the safety feature. https://aiagentnook.com/agent-memory.html https://aiagentnook.com/agent-memory.html Wed, 12 Aug 2026 09:40:00 +0800 The context window as working memory, retrieval-augmented generation for knowledge, external stores for long-term state, and when each earns its complexity. https://aiagentnook.com/agent-tool-use-mcp.html https://aiagentnook.com/agent-tool-use-mcp.html Wed, 12 Aug 2026 09:50:00 +0800 Function calling, writing tool descriptions the model can actually use, and the Model Context Protocol — the open standard connecting agents to external systems. https://aiagentnook.com/multi-agent-systems.html https://aiagentnook.com/multi-agent-systems.html Wed, 12 Aug 2026 10:00:00 +0800 The orchestrator–worker pattern, what parallel subagents buy you, the roughly 15× token cost, and the coordination failures nobody budgets for. https://aiagentnook.com/agent-failure-modes.html https://aiagentnook.com/agent-failure-modes.html Wed, 12 Aug 2026 10:10:00 +0800 Documented AI agent failure modes — direct and indirect prompt injection, the lethal trifecta, runaway loops, and silent reliability drift — and the guardrail that answers each. https://aiagentnook.com/cost-of-running-agents.html https://aiagentnook.com/cost-of-running-agents.html Wed, 12 Aug 2026 10:20:00 +0800 Why agent token bills multiply with every loop iteration — about 4× a chat for one agent, 15× for multi-agent — the hidden costs, and the levers that control them. https://aiagentnook.com/agent-frameworks.html https://aiagentnook.com/agent-frameworks.html Wed, 12 Aug 2026 10:30:00 +0800 What LangGraph and the OpenAI Agents SDK actually add, what the abstraction layer costs you, and when a framework earns its place.