An engineering field guide for AI systems
Learn AI by looking
under the hood.
Start with the automation engineers already trust. Then add models, context, tools, memory and agents—without letting a confident answer pretend it is a system design.
The mental model
A model proposes. The system decides.
A useful AI system has a model in it, but it also has context, policy, tools, state, events and a control loop. We will name each part, build it gradually, and see where it can fail.
Model
generates a proposal
Harness
interprets the proposal
Policy
allows, asks or denies
Tools
perform permitted work
State & events
record what happened
Control loop
continues or stops
A connected curriculum
One application, deeper questions.
Every level grows the same Engineering Troubleshooting Workbench. You should never have to pretend a new buzzword is a new foundation.
Level 1 · Foundations
From Automation to Agentic Systems
A practical journey from rules, workflows and evidence to a small, bounded troubleshooting agent. You will understand every moving part because you build it in public.
Level 2 · Advanced
Agent Systems, MCP and A2A
The good stuff: agent roles, delegation, MCP, A2A, capability boundaries and how to make a multi-agent system understandable instead of merely impressive.
Level 3 · Production
Agents in Production
Observability, Kubernetes, OpenTelemetry, OpAMP, Grafana, Golden Signals, resilience and security for systems that have to earn trust in production.
For people who care about the details
You do not need to know everything to begin.
Most of us have nodded along to a term we did not fully understand. This is a place to stop doing that. Ask plain questions, run the small examples, and follow the evidence.