Learn agentic systems¶
This is the education space. It explains the enduring concepts behind agentic systems without asking you to choose a framework.
How to use this path
Follow the modules in order on your first visit. Each one tells you what you will understand, what you will make, and where to go next. Your progress is stored only in this browser.
The path¶
See the whole system
Model, harness, tools, state, boundaries, and environment.
Outcome: draw the architecture and assign each responsibility.Make one reliable loop
Turn model suggestions into controlled tool calls and a bounded stop.
Outcome: run and break a minimal agent loop.Control context and memory
Decide what enters the prompt, what stays outside, and what can be recalled.
Outcome: explain retrieval, memory, and context growth separately.Connect tools and agents
Use contracts and protocols without confusing interoperability with trust.
Outcome: expose and consume one MCP tool.Design for failure
Retries, approvals, checkpoints, restarts, and durable state.
Outcome: recover without repeating unsafe work.Evaluate behavior
Turn claims into tasks, traces, graders, and comparable evidence.
Outcome: design an evaluation that can prove its claim.Set boundaries and operate
Authority, isolation, observability, deployment, and incident response.
Outcome: define the safe operating envelope for a harness.Explore the wider ecosystem¶
Once the mental model is stable, use the ecosystem map to see where models, protocols, runtimes, sandboxes, evaluation, observability, and governance fit. When you want to implement the ideas, move to Build. When you need product evidence, switch context to Compare.