Post

Production Agentic AI — Engineering for Scale: 30-Day Plan

The third series. After covering individual AI tooling and team-level adoption, this one goes into the production engineering layer: protocols, RAG, advanced agent architectures, security, and scaling. 30 posts, June 10 to July 9.

Production Agentic AI — Engineering for Scale: 30-Day Plan

Two series done. The first covered the tools — Claude Code, GitHub Copilot, Copilot Studio, coding agents — from an individual engineer’s perspective. The second covered what changes at the team level when AI is adopted seriously.

This series goes deeper into the engineering layer that both series touched but didn’t fully cover: building agentic AI systems that work in production at scale.

The timing is deliberate. The 2026 agentic AI landscape has crystallised enough to write about it with specificity. MCP has become the de facto standard for tool connectivity. LangGraph and CrewAI have clear production identities. The LLM pricing collapse has changed what’s economically viable. The EU AI Act is fully in force. And the gap between frontier models has narrowed to the point where model selection is no longer the primary architectural decision.

This series is for engineers who’ve moved past “AI is interesting” and are now asking “how do I build this reliably at scale?”


The Five Arcs

Arc 1 — The 2026 Agentic AI Landscape (Days 1–6)

DayPost
1The Model Wars Are Over — What Model Convergence Means for Engineers
2MCP Explained — The Protocol That Connects AI to Everything
3A2A Protocol — Agent-to-Agent Communication at Enterprise Scale
4LangGraph vs CrewAI — Picking the Right Agent Framework in 2026
5The 13% Problem — Why Enterprise AI Adoption Is Failing
6The LLM Pricing Collapse — How $0.10/Million Tokens Changes Architecture

Arc 2 — RAG and Knowledge Architecture for Agents (Days 7–12)

DayPost
7RAG in Production — Beyond the Basics
8RAG vs Fine-Tuning — The Hybrid Answer in 2026
9Context Window Management in Production Agents
10Vector Databases in 2026 — Which to Use and When
11Chunking Strategies That Actually Work in Production RAG
12Evaluating RAG Pipelines — The Metrics That Matter

Arc 3 — Advanced Agent Architecture Patterns (Days 13–18)

DayPost
13Stateful Agents — Managing State in Production
14Computer Use Agents — The New Agentic Paradigm
15Reasoning Models for Agents — When Thinking Tokens Are Worth It
16Multi-Model Orchestration — SLM + LLM Hybrid Architectures
17Long-Term Memory Patterns for Production Agents
18Tool Orchestration at Scale — Beyond Simple Function Calling

Arc 4 — Security, Safety, and Reliability (Days 19–24)

DayPost
19The 62% Problem — Security Flaws in AI-Generated Code
20Prompt Injection in Production Agents — Attack Patterns and Defences
21Agent Containment — Controlling Blast Radius in Autonomous Systems
22EU AI Act for Engineers — What You Actually Need to Do
23Evaluating Agent Quality at Production Scale
24Observability for Complex Agentic Systems

Arc 5 — Scaling, Frameworks in Depth, and the Road Ahead (Days 25–30)

DayPost
25Scaling Agentic Systems — Cost, Latency, and Reliability
26Building Production Agents with LangGraph — A Hands-On Walkthrough
27CrewAI for Enterprise Multi-Agent Workflows
28The Reasoning Model Revolution — Beyond Next-Token Prediction
29Open-Source vs Commercial Models — The 2026 Decision
30Where Agentic AI Goes Next — The Honest View

Day 1 starts tomorrow. The series runs through July 9.

This post is licensed under CC BY 4.0 by the author.