You will first learn Python and ordinary application engineering. Those skills let you understand the control loop instead of treating a framework as magic. Next come model inputs and outputs, retrieval, evaluation, and access control. Only then do the agent-specific modules add planning, memory, coordination, and durable execution.
Use fake model decisions and sandbox tools while learning runtime behavior. This makes invalid arguments, timeouts, retries, and rejected approvals reproducible without API costs. Later, evaluate actual model decisions separately. Passing fake-client tests does not establish real model quality.
Keep one evolving support-agent project. Start with a fixed workflow, add a bounded tool loop, then persistent state and approval. Each added capability should have a test and an explanation of the problem it solves. All stages are self-paced.
Exercise: write a three-step workflow on paper, then manually simulate a missing document and a rejected action. Record the state after each step.
Check: you can say why the workflow stops and what, if anything, it changed before stopping.
Choose your entry point with evidence
New learners should follow the embedded Python and AI foundations in order. If you completed the AI Engineer path, the shared articles are the same source content; use your prior project evidence to establish readiness rather than repeating the implementation. Progress indicators currently remain specific to the path and browser you use.
Fast-track assessment: demonstrate an authenticated request boundary, tenant-scoped retrieval, validated model output, a held-out evaluation, and a dependency-outage fallback. Then implement a three-step tool loop with a hard stop using only fake decisions. If you cannot explain one of those boundaries, revisit the corresponding foundation module before adding agent autonomy.
The specialization builds one support workflow through task graphs, guarded tools, memory, durable state, planning, framework comparison, MCP, research, coordination, approvals, computer use, evaluation, interruption, and queue operations. Each technical module contains a guided workshop with a worked example or concrete simulation and an independent challenge.
Evidence map: keep tool contracts, state diagrams, memory scope tests, proposal and approval records, operation IDs, failure traces, and budget reports in your project. The final capstone combines these artifacts. Adding more tools or agents is not a completion criterion; demonstrating useful behavior inside enforced boundaries is.
Worked placement decision: a developer who has built a chatbot but has never implemented a timeout or authorized retrieval should take the shared gateway and isolation labs. Familiarity with an SDK does not establish those prerequisites. A developer who can already reproduce the AI capstone may move directly to the task-graph lab after reading the agent orientation.
Checkpoint: identify which parts of your system are deterministic application code, which are model decisions, and which require human authority. You should be able to stop the system, inspect its persisted state, and explain what it is allowed to do next without asking the model to reconstruct the truth from a transcript.