Build Reliable AI Agents with Jev, LLMs & MCP
Jev is emerging as a new piece of the agentic AI stack. Learn where it fits alongside LLMs and build a complete AI agent using Jev, Python, Tools and MCP.
Your AI agent probably doesn't need an LLM for every decision.
Routing a request, choosing a tool, classifying an input, scoring a result, or deciding whether to escalate are often decision-making tasks rather than generation tasks.
Yet many AI agents still rely on LLMs for almost every step, adding unnecessary token costs, latency, and complexity.
What if your agent could use a faster, more focused model for routine decisions and reserve LLMs for tasks that genuinely require generation or deeper reasoning?
In this 4-hour live, hands-on workshop, Python author and software architect Jaime Buelta will show you how to build reliable AI agents using Jev + LLMs + Python + Tools + MCP.
You'll explore how Jev, a System 1 model designed for structured decision-making, can work alongside LLMs to create more efficient, controlled, and reliable agent workflows.
You'll go beyond the concepts and build a working agentic application, complete with model integration, tool use, MCP servers, decision logic, and reliability checks.
What You'll Learn
Jev vs. LLMs: Understand when to use a decision model and when an LLM is needed.
Reduce unnecessary LLM calls: Use Jev for routing, classification, tool selection, scoring, and other structured decisions.
Build AI agents with Python: Work with agent loops, context, local and hosted models, and APIs.
Integrate tools and MCP: Give agents access to external capabilities and control how they use them.
Work with Claude Code or Codex: Use coding agents, MCP servers, and skills as part of a structured development workflow.
Improve agent reliability: Handle failures, inspect agent behaviour, and introduce verification and escalation logic.
Build a complete application: Combine Jev, LLMs, Python, MCP, and tools into a working agentic system.
Workshop Agenda
Module 1: Understanding AI Agents, LLMs, Jev & MCP
Understand the components of a modern AI agent and how they work together.
What an AI agent is and how it works
LLMs vs. System 1 / Jev-style models
Where Jev fits in decision-making workflows
Local vs. hosted models
Context, memory, and agent loops
MCP and other approaches to tool integration
General-purpose vs. specialized agents
Module 2: Agentic Coding That Works
Learn to use coding agents as part of a structured engineering process.
Work with Claude Code or Codex
Extend coding agents using MCP servers and skills
Move beyond vibe coding to structured, reviewable development
Use specifications, tools, and reusable workflows to improve consistency
Module 3: Building AI Agents with Jev & MCP
Start building the core components of a hybrid agent architecture.
Understand the components that make up an agent
Connect hosted and local LLMs through APIs
Give your agent tools and control how they're used
Use Jev for decisions, categorization, and routing
Create and connect an MCP server
Combine decision-making and generation within a practical workflow
Module 4: Capstone – Build a Reliable Agentic Application
Bring everything together into a complete working project.
Integrate Jev, LLMs, Python, MCP, and external tools
Use different models for different tasks
Handle failures and improve reliability
Inspect agent behaviour and refine decision logic
Experiment with and extend the application
Your Agent Reliability Kit
Every workshop ticket is planned to include a reusable Agent Reliability Kit, designed to help you apply what you've learned to your own projects.
Reusable Agent Skills:
Decision Router SKILL.md
Model Selector SKILL.md
Tool Gate SKILL.md
Confidence & Escalation SKILL.md
Output Verifier SKILL.md
Agent Trace Reviewer SKILL.md
Developer Resources:
Python starter project
MCP starter implementation
Agent reliability checklist
These resources are designed to help you structure agent decisions, control tool use, inspect behaviour, and improve reliability beyond the workshop.
Who Should Attend?
This workshop is ideal for:
AI and ML Engineers building agentic applications
Python and Software Developers exploring AI agent development
AI Agent Builders looking to reduce unnecessary model calls
Claude Code and Codex Users interested in MCP, tools, and agent architecture
LangGraph and MCP Developers exploring hybrid agent systems
Software Architects and Tech Leads designing more efficient AI workflows
A basic understanding of Python, APIs, and LLMs will be helpful. No prior experience with Jev is required.
What's Included in Your Ticket?
4-hour live, hands-on workshop with Jaime Buelta
Live Q&A with the instructor
Complete workshop recording
Code examples and workshop resources
Capstone project and supporting materials
Agent Reliability Kit (planned bonus)
Certificate of Completion
By the End of the Workshop
You'll have built an agentic application that combines Jev + LLMs + Python + Tools + MCP, with practical experience in controlling tool use, making structured decisions, handling failures, and improving reliability.
More importantly, you'll understand how to design agents that don't rely on an LLM for every step, helping you reduce unnecessary model calls and build more efficient systems.
Don't pay for reasoning when you only need a decision.
Learn how to build agents that use the right model for the right task.
Meet Your Instructor
Jaime Buelta is a software architect, developer, author, and long-time Python practitioner.
His work spans Python development, software architecture, automation, microservices, and building scalable software systems. He is the author of Packt titles including Python Automation Cookbook and Python Architecture Patterns. His latest work around Python automation also covers AI models, MCP integrations, and the creation of AI agents.
Throughout this workshop, Jaime will bring an engineering-first perspective to AI agents focusing not just on what agents can do, but how their underlying systems should be designed, inspected, and improved.