Build a Morning Intelligence and Trend-Watching Agent
Build a bounded morning intelligence agent that preserves source evidence, ranks developments without overclaiming, exposes failures, and produces one review-ready brief.
Define the Briefing Decision and V0 Boundary
Choose one audience, topic band, decision, cadence, destination, and explicit boundary for a useful morning brief.
Build the Source Policy and Watchlist
Create a small source register with approved lanes, freshness rules, exclusions, and primary-source verification requirements.
Design Bounded Collection and Run Controls
Set exact queries, caps, cost controls, secret handling, stop rules, and a run manifest before collecting live evidence.
Preserve Evidence and Normalize Unknowns
Save raw records first, create stable identities, normalize dates and sources, deduplicate safely, and preserve unknown states.
Rank Developments Without Overclaiming
Separate facts, derived values, interpretation, confidence, and recommendations in a versioned evidence-linked rubric.
Design the Morning Report and Human Review
Turn one normalized run into a concise brief with reconciled counts, evidence, unknowns, recommendations, and review boundaries.
Test, Schedule, and Launch the Agent
Run deterministic fixtures, define the approval-dependent canary, schedule one bounded report, and grade the complete launch pack.
What you will work towards
Work through this course to prepare a bounded morning intelligence report and launch pack. The curriculum takes you through a briefing charter, source policy, collection caps, evidence contract, ranking rubric and fixture suite.
Start with a named reader, a decision and an approved watchlist. Use synthetic information while learning when permission to use live records is unresolved.
A practical case to carry through the lessons
Several stories repeating the same announcement are not several independent developments. Preserve source dates and deduplicate the event before deciding that it is a trend.
Test stale items, missing dates, blocked sources and a run with no relevant developments. Keep an empty result distinct from a failed collection.
How to know the work is ready for review
Before calling the exercise complete, check that the reader can separate observation from interpretation and the operator can see, pause and recover a failed run.