🤖 Agent Builder Path

• Anatomy of an Agent: Model, Tools, Memory, and the Loop (7 min) — An AI agent is a system that uses a language model (LLM) as its reasoning core and wraps it with four components that t…
• The Agent Loop: Plan, Act, Observe, Repeat (8 min) — An AI agent operates in a continuous reasoning loop often called the ReAct loop (Reasoning + Acting). Each iteration ha…
• Designing Good Tools for Agents (8 min) — When an LLM-based agent decides which tool to call, it relies entirely on three things: the tool's name, its descriptio…
• MCP: The USB Port for AI Tools (7 min) — MCP (Model Context Protocol) is an open standard, introduced by Anthropic in late 2024, that defines how AI agents comm…
• Agent Memory: Short-Term Context, Long-Term Storage, and Staying Coherent (8 min) — Every LLM-based agent has a fixed context window — the maximum number of tokens it can 'see' at once (e.g., 128k tokens…
• Evaluating Agents: Measuring Task Success and Catching Silent Failures (8 min) — Evaluating agents is fundamentally different from evaluating single-turn LLM responses. A single-turn eval checks one o…
• Shipping Agents Safely: Guardrails, Approvals & Injection Defense (9 min) — When you deploy an AI agent in production, four safety layers matter most. **Guardrails** are hard rules checked before…

Anatomy of an Agent: Model, Tools, Memory, and the Loop

An AI agent is a system that uses a language model (LLM) as its reasoning core and wraps it with four components that together enable autonomous, multi-step task completion. **Model** — the LLM that reads context and decides the next action (e.g., call a tool, ask a clarifying q…

The Agent Loop: Plan, Act, Observe, Repeat

An AI agent operates in a continuous reasoning loop often called the ReAct loop (Reasoning + Acting). Each iteration has four phases: Plan (the model decides what action to take next, given the goal and memory so far), Act (it calls a tool or produces output), Observe (the tool …

Designing Good Tools for Agents

When an LLM-based agent decides which tool to call, it relies entirely on three things: the tool's name, its description, and its parameter schema. These are not just documentation — they are the agent's decision-making inputs at runtime. **Names** should be short, verb-noun sna…

MCP: The USB Port for AI Tools

MCP (Model Context Protocol) is an open standard, introduced by Anthropic in late 2024, that defines how AI agents communicate with external tools and data sources. Think of it as a universal adapter layer: instead of writing bespoke integration code for every tool an agent migh…

Agent Memory: Short-Term Context, Long-Term Storage, and Staying Coherent

Every LLM-based agent has a fixed context window — the maximum number of tokens it can 'see' at once (e.g., 128k tokens for GPT-4o, 1M for Gemini 1.5 Pro). Everything inside that window is short-term memory: the system prompt, conversation history, tool results, and the current …

Evaluating Agents: Measuring Task Success and Catching Silent Failures

Evaluating agents is fundamentally different from evaluating single-turn LLM responses. A single-turn eval checks one output against one expected answer. An agent eval must assess a multi-step trajectory — the sequence of actions, tool calls, and intermediate states the agent pr…

Shipping Agents Safely: Guardrails, Approvals & Injection Defense

When you deploy an AI agent in production, four safety layers matter most. **Guardrails** are hard rules checked before or after every action — e.g., block any tool call that would delete data, or reject outputs containing PII. They can be implemented as code validators, regex f…