Course Pass 10: Agents

Go agent SDK on their engine’s tool calls, RAG over their docs, eval runner with judge and A/B, usage policy at the gateway, agent chart; optional post-training. About 5 weeks at 10 to 12 hours a week.

Part of: the course. Needs: Pass 9.

Gate: MS-P10 = MS-agent (+ optional MS-C2).

Stages, chapter links, and check columns are in path.tsv. Use practice/bin/ol learn course-p10-p10-agents to see progress.

Stages

Generated from path.tsv. Track progress locally with practice/bin/ol learn.

Stage	Read	Done when
1	Types, OpenAI-compatible provider, retry wrapper Types, OpenAI-compatible provider, retry wrapper	Your implementation of ag.01 passes its course checks.
2	Tools, registry, JSON Schema argument validation Tools, registry, JSON Schema argument validation	Your implementation of ag.02 passes its course checks.
3	Agent loop Agent loop	Your implementation of ag.03 passes its course checks.
4	Tool gate, SQL safety gate, prompt-injection suite Tool gate, SQL safety gate, prompt-injection suite	Your implementation of ag.04 passes its course checks.
5	Durable agent runs (AgentRun) Durable agent runs (AgentRun)	Your implementation of ag.05 passes its course checks.
6	RAG ingest: crawler with SSRF allowlist, chunker, embedder, store RAG ingest: crawler with SSRF allowlist, chunker, embedder, store	Your implementation of ag.06 passes its course checks.
7	Retrieval: BM25, flat and IVF vectors, RRF, MMR Retrieval: BM25, flat and IVF vectors, RRF, MMR	Your implementation of ag.07 passes its course checks.
8	Context assembly, citations, search_docs tool Context assembly, citations, searchdocs tool	Your implementation of ag.08 passes its course checks.
9	Eval runner Eval runner	Your implementation of ag.09 passes its course checks.
10	Agent evaluation scorers Agent evaluation scorers	Your implementation of ag.10 passes its course checks.
11	LLM judge with position control LLM judge with position control	Your implementation of ag.11 passes its course checks.
12	Paired A/B experiments Paired A/B experiments	Your implementation of ag.12 passes its course checks.
13	Usage policy at the gateway Usage policy at the gateway	Your implementation of ethics.05 passes its course checks.
14	Usage policy enforcement Usage policy enforcement	Your implementation of gw.08 passes its course checks.
15	Agent image and chart Agent image and chart	Your implementation of dep.07 passes its course checks.
16	Agent evals as tests (R9) Agent evals as tests (R9)	Your implementation of craft.23 passes its course checks.
17	Milestone agent Pass 10 milestones	The MS-agent gate passes for your system.
18	Milestone P10 Pass 10 milestones	The MS-P10 gate passes for your system.
19	Optimization problem set, part b: constrained optimization, duality, and DPO Optimization problem set, part b: constrained optimization, duality, and DPO	You complete the S-M10b solve set and its check passes.
20	Supervised fine-tuning: chat rendering, assistant-only loss, packing Supervised fine-tuning: chat rendering, assistant-only loss, packing	Your implementation of L12.1 passes its course checks.
21	Direct preference optimization and IPO Direct preference optimization and IPO	Your implementation of L12.2 passes its course checks.
22	Group relative policy optimization with verifiable rewards Group relative policy optimization with verifiable rewards	Your implementation of L12.3 passes its course checks.
23	Distillation with forward and reverse KL Distillation with forward and reverse KL	Your implementation of L12.4 passes its course checks.
24	Capstone 2: post-trained TinyStories model Capstone 2: post-trained TinyStories model	Your implementation of C2 passes its course checks.
25	Milestone C2 Pass 10 milestones	The MS-C2 gate passes for your system.