- Accueil
- Montréal 2027
- Sessions
This comprehensive 2-day workshop equips engineering leaders, architects, and developers with the practical patterns needed to move beyond basic IDE prompting into production-grade, fullstack agentic software delivery.
Whether your enterprise stack is built on Java, .NET, Python, PHP, or TypeScript, the architecture of reliable AI systems remains identical. You will learn how to transition from managing individual coding assistants to orchestrating a fully realized Agentic Software Factory, leveraging spec-driven development, cross-platform agent kits, and portable execution models to build secure, verifiable multi-agent teams.
We will tackle the exact challenges keeping engineering teams up at night: preventing context poisoning, sandboxing agent execution, defending against prompt injection, and enforcing deterministic verification loops before code ever hits production. Through hands-on labs, you will build autonomous multi-agent pipelines, integrate Model Context Protocol (MCP) toolchains, and orchestrate real-time event-driven workflows that safely execute long-horizon tasks.
While code examples and real-time UI streaming labs utilize TypeScript and Node.js, the core multi-platform orchestration models, MCP protocols, and security guardrails taught are designed to be deployed across diverse enterprise backend and frontend environments.
Intro - The Shift to the Agentic Software Factory
- Moving from Fragile Prompts to Deterministic System Architecture
- Deconstructing Work: How to split software delivery tasks effectively across agent teams
- Polyglot Integration: Cross-platform SDK portability for Java, Python, .NET, PHP, and TS
- Context Architecture: Project memory, workspace state, and avoiding context pollution
- Agent Managers and Teamwork Structures (Structuring tasks via AGENTS.md and SKILL.md)
Part 1 - Spec-Driven Development & Context Engineering
- Bridging Human Intent to Agent Execution via Structured Specs
- Context Window Economics and Progressive Disclosure
- Enforcing Guardrails: Ensuring agents code against strict technical contracts, not guesses
- Lab: Setting Up Spec-Driven Workflows and Guardrails
Part 2 - Managing Multi-Agent Teams & The Deterministic Loop
- Orchestrating Cross-Platform Agent Teams
- When to Change Your Agent Orchestration: Scaling from linear pipelines to complex collaborative topologies
- Model-as-Judge Best Practices: Using distinct AI vendors to cross-verify execution and eliminate single-model hallucinations
- The Plan – Execute – Verify Architecture: Enforcing rigorous verification loops
- Lab: Creating Your First Collaborative Multi-Agent Team
Part 3 - MCP and Agent-to-Agent Communication
- Building Tool Interoperability with the Model Context Protocol (MCP)
- External Data Grounding, Custom Tools, and Secure Enterprise APIs
- The Agent-to-Agent (A2A) Protocol: Discovery, Agent Cards, and Remote Delegation
- Lab: Constructing an Enterprise Team with Secure Tool Access via MCP and A2A
Part 4 - Real-Time Multimodal Streaming & Reactive UI
- Developing Real-Time Audio/Video Pipelines (WebRTC & WebSockets)
- Stream-Driven UI and Live Agent State Synchronization
- Handling Latency, Interruptions, and Live Transcription in Agent Loops
Part 5 - Asynchronous Execution & Long-Horizon Workflows
- Event-Driven Agent Teams and Heartbeat Loops
- Persistent Memory and State Management for Long-Running Tasks
- Lab: Building an Asynchronous Pipeline Where Intents Spawn Long-Horizon Tasks (Ticketing, Code Synthesis, and Automated Research)
Part 6 - Enterprise Security, Guardrails & Autonomous Operations
- Agentic Security Deep Dive: Defending Against Indirect Prompt Injection
- Sandboxing & Least Privilege: Restricting file access and tool execution
- Performance Evals, Budget Guards, and Automated Compliance Checks
- Lab: Producing a Fully Autonomous, Guardrailed Agent Team That Plans, Builds, and Verifies Safely Until Done
Participant should have:
- Intermediate skills
- Basic knowledge of backend and frontend development
Duration:
- 2 days
- 9:00 am to 5:00 pm
- 1 hour lunch break included at the hotel's restaurant
- 15 min coffee break every morning and afternoon
Gerard Sans
Google Developer Expert; former AWS Amplify team
Gerard Sans is a Computer Science Engineer and Google Developer Expert in AI, Cloud, and Angular. Formerly an AWS Amplify Developer Advocate, he now leads Axiom Professional Trainings and organises the NextAI London community. Passionate about the future of AI, Gerard enjoys sharing his knowledge through talks, workshops, and mentoring.
Read More