Exercises — Module 09: Frontier topics

9.1 — Literature review

Pick one frontier topic from Module 09 (long-horizon agents, computer use, memory systems, multi-agent coordination, eval at scale, etc.). Write a three-page literature review of primary sources (papers, technical blog posts by the people doing the work). No secondhand summaries.

Deliverable: three pages with a citation list.

9.2 — Weekend attempt

From your chosen topic, pick an open sub-problem that looks tractable in a weekend. Try. Expect to fail. Write up what you tried, what you learned, and what you would try next if you had another week.

Deliverable: the attempt + the post-mortem.

9.3 — Interview

Find someone who works on agents at a different organization. Interview them about the hardest problems they currently face. Compare to Module 09. Update your mental model of the field. Save the “before” and “after” notes.

Deliverable: the interview notes and the delta.

9.4 — Model cascade

Implement a cheap-then-escalate cascade: try a small model first, fall back to a larger model if a confidence check fails. Measure cost and quality on a 30-case eval. Does it beat the large model alone at equivalent quality?

Deliverable: results.

9.5 — Persistent memory

Add a persistent memory file (simple JSON) that your harness reads at start and writes to at end. Run it across 5 sessions on related tasks. Does session N benefit from sessions 1..N-1? How did memory drift?

Deliverable: traces + analysis.