AI-Assisted Coding | Session 3: Agentic AI — Agents, Workflows and Capstone

AI-Assisted Coding | Session 3: Agentic AI — Agents, Workflows and Capstone

We offer a 50% discount to students; please complete this form to avail the discount.

How do you let an AI agent plan, edit and test code while ensuring every action stays bounded, traced and approved?

This session explores agentic coding workflows, covering agent loops, state, stopping conditions, typed tools with validated schemas and retries, routing, guardrails, tracing and human approval.

You will learn to model plan, inspect, edit, test and review as a stateful LangGraph workflow. The session also covers task suites, pass rates, sandboxing, unsafe commands, dependency risk, repository prompt injection, trace metrics and cost budgets.

The programme concludes with a capstone project that combines repository retrieval, MCP tools and a controlled workflow into a code-generation assistant, verified through functional, retrieval, permission and adversarial tests.

This session is the third and final session of the three-session professional training programme:

• Session 1 — Code Generation: Safe Prompts and Verified Change

• Session 2 — Repository Context: RAG and MCP

• Session 3 — Agentic AI: Agents, Workflows and Capstone

REGISTRATION FOR ALL 3 SESSIONS IS 150$ ONLY.

What You Will Learn

• Agent loops, state and stopping conditions

• Typed tools with validated schemas and retries

• Routing, guardrails and tracing

• Human approval controls

• Stateful LangGraph workflows

• Plan, inspect, edit, test and review

• Task suites and pass rates

• Sandboxing and unsafe-command prevention

• Dependency risk and repository prompt injection

• Trace metrics and cost budgets

• Capstone architecture and acceptance testing

• Operational handover documentation

• Functional, retrieval, permission and adversarial testing

Why This Matters

An agentic coding system can perform multiple actions across a repository, so it's essential to control what it can access, execute, and modify.

This session brings together the engineering discipline developed throughout the programme, applying bounded tasks, controlled context and verified output to an integrated coding assistant.

Ideal For

• Software engineers

• Technical leads

• QA engineers

• Software architects

• Engineers building AI agents and automated coding workflows

• Final-year and postgraduate learners with programming and Git experience

Basic programming ability, preferably in Python, and basic Git and command-line experience are required.

Training Format

• 4-hour live online training session

• Concise teaching blocks

• Four live demonstrations:

  • Demo 9: Build an issue-triage agent with typed tools, guardrails, tracing and a human approval step.

  • Demo 10: Model plan, inspect, edit, test and review as a stateful LangGraph workflow.

  • Demo 11: Run evaluation and prompt-injection tests and inspect the security, trace and cost evidence.

  • Demo 12: Build a code-generation assistant combining repository retrieval, MCP tools and a controlled workflow, verified with functional, retrieval, permission and adversarial tests.

• A ten-minute break and five-minute quiz exactly at the two-hour mark

• Capstone completion and operational handover pack

Tools and Technologies

Claude Code or Codex, Python, Git, pytest, Qdrant, the official MCP Python SDK, LangChain and LangGraph.

Capstone Outcome

A code-generation assistant that combines repository retrieval, MCP tools and a controlled agentic workflow, supported by verification evidence and an operational handover pack.

This session is Session 3 and the final session of the AI-Assisted Coding and Code Generation Programme.