Agent Engineering
From using agents to building them
Move beyond using and configuring AI coding tools. Understand the core mechanics of the agent runtime and build agent applications that are observable, evaluable, and safe to operate.
What this track is for
The Growth Route explains how to become progressively better at AI Coding. This specialization answers a different set of questions: Why can an agent keep acting? How does it choose tools, maintain state, operate in a real environment, and recover from failure? What turns a working demo into a safe, reliable, governable engineering system?
The track reuses the site’s existing material on context, MCP, skills, multi-agent systems, gateways, and evaluation while gradually adding implementation courses for agent runtimes and coding agents. Every article still has one source and remains part of the original four growth stages.
Start with one model call and implement a minimal agent loop.
Let the agent safely read, modify, execute, and verify real projects.
Add state, recovery, observability, evaluation, cost, and safety boundaries.
Four learning modules
How agents work
Understand the feedback loop between model, runtime, and tools without depending on any one framework.
- Planned From one model call to an agent loop
- Planned Tool calling, event streams, and stopping conditions
- Foundation Understand tokens and task cost through the feedback loop
Core coding-agent systems
Build a minimal coding agent that understands a repository, edits files, executes commands, runs tests, and continues from the evidence.
- Foundation Repository search, context selection, and compression
- Planned Files, terminal, Git, and the testing loop
- Planned Sessions, state, memory, and workspaces
- Planned Approvals, sandboxes, and error recovery
Agent capabilities and orchestration
Learn how capabilities connect, package, and compose—and when deterministic workflows, multiple agents, or durable execution are warranted.
- Available MCP, tools, and external systems
- Available From prompts to agent skills
- Available Engineering rules, skills, and plugins
- Available Complex tasks and multi-agent workflows
- Planned Choosing an agent SDK or orchestration framework
Production and governance
Move from an agent that sometimes works to one that is observable, evaluable, recoverable, and bounded by cost and security controls.
- Available Model gateways, routing, permissions, and auditing
- Available Benchmarks and real engineering evaluation
- Planned Tracing, checkpoints, and long-running recovery
- Planned Agent security, deployment, and versioning
Completion standard
This track is not complete when you have learned a certain number of frameworks. Its capstone is a minimal but complete coding agent: it can read a real repository, choose tools, modify code, run verification, request human approval before high-risk actions, and preserve an execution trace that can be reviewed and evaluated.