The five types of agent memory and why memory is a lifecycle rather than a database, plus building a deliberately stateless customer support agent with a live order-status tool.
Tag: Generativeai
Summary of Today’s Gen AI MCP Session: Your First Server – Tools, the Inspector and Claude Desktop (27 Oct 2026)
Writing a working MCP server with FastMCP and @mcp.tool(), testing it in the MCP Inspector, wiring it into Claude Desktop (and the config errors that bite), plus versioning and backward compatibility.
Summary of Today’s Gen AI MCP Session: Pipes and Protocols – MCP Architecture and Your Dev Environment (26 Oct 2026)
MCP architecture restated as host, client and server; stdio vs Streamable HTTP transports; what the SDK exposes on each side; and the full Python and Node setup needed before writing your first MCP server.
Summary of Today’s Gen AI MCP Session: USB-C for AI – Booking a Real Table with Claude (24 Oct 2026)
A live demo of an agent booking a restaurant table through an MCP connector, plus the vocabulary behind it: host, client and server, what a server exposes, local vs remote transports, and why businesses are adopting MCP.
Summary of Today’s Gen AI Foundation Session: From Scholar to Assistant – Tool Calling and MCP (22 Oct 2026)
How a pretrained model becomes an assistant through fine-tuning, why an LLM can name a tool but never run one, the vendor lock-in problem with tool definitions, and what the Model Context Protocol solves.
Summary of Today’s Gen AI Foundation Session: Why “Bank” Means River – RNNs, Transformers and Embeddings (15 Oct 2026)
Why plain deep learning fails on language, how RNNs added memory, how Transformers and attention replaced them, what embeddings and tokens really are, the full prompt pipeline, and how to size a model to your GPU.
Summary of Today’s Gen AI Foundation Session: From a Straight Line to a Deep Network – How AI Actually Learns (14 Oct 2026)
The evolution from linear regression to deep learning: machine learning vs traditional programming, what sits inside a neuron, how RLHF fine-tunes a model, what makes a network deep, and why GPUs took over.
Summary of Today’s Agentic AI Session: A Human in the Loop – Using can_use_tool in the Claude Agent SDK (13 Oct 2026)
How the can_use_tool callback works in the Claude Agent SDK: where it sits in the permission evaluation flow, its three arguments, a working example, five ways to respond, which permission modes call it, and the gotchas.
Summary of Today’s Gen AI Foundation Session: Auto-Complete at Scale – What an LLM Actually Is (13 Oct 2026)
First foundation session: the difference between an application and a model, how next-token prediction works, tokens and pricing, training vs inference, and why the neural-network analogy holds.
Summary of Today’s Gen AI Classroom: It Can Touch Your Disk Now – Tools and Permissions (7 Oct 2026)
Building the Downloads Folder Agent with built-in tools, parsing tool-use and tool-result blocks, and the permission system: allowed_tools vs disallowed_tools, permission modes, and a supervised can_use_tool workflow.
Summary of Today’s Gen AI Classroom: Who Remembers You? – Sessions, State and Scaling (6 Oct 2026)
Giving the tutor agent memory with ClaudeSDKClient and session IDs, why local session storage breaks in production, shared session stores for horizontal scaling, and mapping users to sessions.
Summary of Today’s Gen AI Classroom: Context in a File – CLAUDE.md and Coding Agents (5 Oct 2026)
Project-level agent behaviour with CLAUDE.md, controlling settings_sources, how the Claude SDK uses the CLI as a harness, installing and using Claude Code, and what coding agents change about testing and DevOps work.
Summary of Today’s Gen AI Project: LangGraph Memory, Threads and Summarization (3 Oct 2026)
Project session on LangGraph: proving a stateless agent forgets, adding checkpointers and thread IDs, inspecting state snapshots, persistent storage, and building a summarization node to control token cost.
Summary of Today’s Gen AI Classroom: Three Tones, One Agent – Driving Behaviour from a Dictionary (1 Oct 2026)
Making the tutor agent’s teaching style configurable: tones in a dictionary, .get() with a safe fallback, a reusable options helper with **kwargs, cost control, and why this isn’t an agent yet.
Summary of Today’s Gen AI Classroom: One Prompt, Many Personas – System Prompts and the Tutor Agent (30 Sep 2026)
System prompts across Claude, OpenAI and Gemini, designing a tutor agent with three teaching modes, query vs client, and the full uv project setup for the Claude Agent SDK.
Summary of Today’s Gen AI Classroom: Agent, Runner, Tool – Building with the OpenAI Agents SDK (29 Sep 2026)
Building an agent in pure Python with the OpenAI Agents SDK: Agent, Runner and Tool, custom tools with function_tool, SQLite session memory, and how it differs from the Claude Agent SDK.
Summary of Today’s Gen AI Classroom: Let the Framework Do the Plumbing – The Claude Agent SDK (28 Sep 2026)
Moving from low-level API calls to the Claude Agent SDK: async basics, query vs client, reading the message hierarchy, model cost control, tool permissions, and where MCP fits.
Summary of Today’s Gen AI Classroom: The Model Forgets, and It Can’t Press the Button – Memory and Tool Calling (24 Sep 2026)
Why raw API calls have no memory, how to parse Claude and OpenAI responses and track token cost, how tool calling really works, and the two-week roadmap from simple agents to MCP, A2A and RAG.
Summary of Today’s Gen AI Classroom: Your First API Call – Talking to Claude and GPT in Code (23 Sep 2026)
Setting up a Python project with UV, managing API keys in .env, and making your first direct calls to the Claude and OpenAI APIs — plus what the credit errors mean.
Summary of Today’s Gen AI Classroom: From Brain to Hands – How LLMs Become AI Agents (22 Sep 2026)
How an LLM works, what it can’t do on its own, and how an AI agent fills the gaps with tools and memory. Plus why verification is the new bottleneck in the agentic era.