AJ Learning Hub

Advanced · 6 lessons · about 135 min

Advanced: Agents in Production

Build agents that hold up after launch: the agent loop and its harness, multi-agent orchestration, MCP and A2A, context and memory, evals and tracing, and the security and cost controls that keep a fooled or runaway agent from doing damage.

By the end you'll be able to

  • ✓Write an agent loop with tool calls, error handling and a hard stop, and decide when a fixed workflow is the better design.
  • ✓Choose an orchestration pattern for a multi-agent task and estimate its token, cost and latency impact before building it.
  • ✓Build and test a minimal MCP server, and explain what the stateless 2026-07-28 revision and AAIF governance change for you.
  • ✓Keep long-running agents inside a context budget with prompt caching, compaction, scoped memory and just-in-time retrieval.
  • ✓Ship agents behind eval gates, traces, least-privilege tools, human approval for risky actions and hard spend limits.
0 of 6 lessons0%
  1. ✓1.AI agents and harnessesWhat makes a system an agent, how the tool-calling loop actually runs, why the harness around the model decides whether it works, and when you should not build an agent at all.20m
  2. ✓2.Multi-agent orchestrationWhen splitting work across subagents pays off, the orchestration patterns that recur in practice, how to estimate the token and latency bill, and the coordination failures to design against.22m
  3. ✓3.MCP and A2AHow the Model Context Protocol connects agents to tools and data, what the stateless 2026-07-28 revision changed, how to build and test a minimal server, and where A2A and the Agentic AI Foundation fit.24m
  4. ✓4.Context engineering and memoryTreat the context window as a budget: what to put in it, how prompt caching cuts the cost of resending it, how long-running agents stay inside it, and how to give agents memory without giving them stale or poisoned facts.22m
  5. ✓5.Evals and observabilityTest prompts and agents the way you test code: success criteria, graded eval sets, LLM-as-judge done carefully, regression gates in CI, and traces that let you debug a production run step by step.22m
  6. ✓6.Security and productionDesign agents that stay safe when they are fooled: prompt injection as an architecture problem, least-privilege tools and guardrails, human approval for risky actions, and the deployment and cost controls that keep a production agent predictable.24m