How LLM Agents Actually Work.
A magazine-style, deeply visual reference for building real agentic systems — from the autonomy spectrum to a live evaluator-optimizer loop you can run.
What's inside
A self-contained tour designed to take you from intuition to working code.
Six chapters. Build up. Then build it.
Foundations
The autonomy spectrum and the augmented LLM — what an 'agent' actually is.
Agentic Patterns
Prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer, ReAct.
Controlling Generation
Guidelines, writing profiles, few-shot, structured outputs — the levers that actually work.
The AI Engineering Stack
Context · Tools · Orchestration · Reasoning · Evaluation. Plus the four real constraints.
The PM Spec Framework
Nine questions every PM should answer before shipping an agent flow. Spec the seams, not the model.
PM Design Patterns
Six PM-shaped recipes — Confidence-Gated Escalation, Critic Rubric, Voice-Anchored Generation, and more — each mapped to flow components and spec slots.
Case Studies × 2
The LinkedIn Writer multi-agent system AND the AI PM Briefing Fan-Out + Reflection pipeline. Both pulled apart.
Live Demo
Run the wizard. Watch the audit log. Feel the loop.
Most agent tutorials are vibes. This one shows the wires.
Every diagram, pattern, and control mechanism here maps to real, production-tested ideas — distilled from Anthropic's 'Building Effective Agents', Paul Iusztin's writing, and the LinkedIn Writer system that ships in the sister project 'Agentic AI Creator'.