Building AI Agents

Cui îi este dedicat cursul?

Software engineers, Tech Leads and Architects who have completed AI Introduction & Integration- or have equivalent knowledge of LLMs, Prompt & Context Engineering, ideally also about RAG (Retrieval Augmented Generation)

Cunoștințe și abilități inițiale

- Comfortable with the Terminal / CLI
- Node.js 20+ and an Anthropic API key (personal or company-provided) for the hands-on
exercises

Prezentarea cursului

A hands-on, two-part workshop for engineers who want to build AI agents - from a working PoC, to production-ready agents with advanced tool usage and persistent memory.

Continuation of: AI Introduction & Integration Basic and Advanced trainings (LLMs, Prompt & Context Engineering, RAG & vector DBs, AI tools, agents primer).

Ce subiecte abordează cursul

Part 1 - From Zero to a Working Agent
- Anatomy of an agent: prompt, context, tools, memory, planner, executor - the analyze → plan → execute loop made concrete
- The tool-use loop in detail: how the agent receives tool definitions, picks a tool, executes it, feeds results back, and decides when to stop
- Building a PoC agent end-to-end: TypeScript + Anthropic SDK, file tools, a sandboxed workspace, a REPL with slash commands
- Execution environment - cloud vs on-prem: where the agent will runs:
- Cloud: simpler to deploy, faster to start, limited / no access to internal APIs
- On-prem: more complex to deploy, partial-to-full access to the company's internal systems and APIs
- Working effectively: managing context, prompting for agents, human-in-the-loop discipline, deterministic artifacts over chat
- Hands-on: build the PoC from scratch - Anthropic SDK + tool-use loop + sandboxed file tools - running locally end-to-end

Part 2 - Advanced: Tools, Memory, and Production
- Advanced tool usage: custom tools beyond file I/O - wrapping internal APIs, system commands, RAG retrieval, structured output, parallel tools
- Connecting to internal infra (the on-prem story): secure tool wrappers, secrets and
credentials, scope and RBAC, audit trails - the practical work of making an agent useful inside a
company
- Persistent memory: short-term conversation memory vs long-term knowledge; what to store,
where (file / DB / vector store), how to retrieve, how to keep it bounded
- Claude Code as the builder's IDE: scaffolding, iterating on, debugging, and reviewing your own
agent code with Claude Code - the agent that helps you build agents
- From PoC to production-ready: evaluation, observability, cost control, safety, bounding
autonomy, packaging the agent as a CLI or a service
- Hands-on: evolve the PoC - add a custom tool that calls an internal API, plus persistent memory
across sessions

Ce abilități se dobândesc în urmă cursului

Outcome: attendees leave able to build an AI agent from scratch - understand the moving parts, ship a working PoC, then evolve it with custom tools and persistent memory; and decide where it will run (cloud or on-prem), given their company's tools and constraints.