Choose a prompting technique by task and failure risk
A focus timer must size the block and the break to the work
An agent architecture must fit the task and hold up in production
Choose a model from the job, constraints, and required checks
Agent memory must recall the right thing at the right moment
Design experiments that separate effects from noise
A build loop must plan first and verify before claiming done
Follow how text becomes embeddings, retrieved evidence, and a grounded answer
A graph must hold its invariants and answer the queries you run
Graph retrieval must build the graph before the question arrives
Assemble an agent stack, then inspect its failure risks
Every agent makes the same core decisions; their location varies
Enabling multi-agent cooperation in same repo
Every tool sits in one layer, and the layer decides what it may do
Learn how embeddings, vector databases, retrieval, and RAG fit together
Watch two agents talk it through, then run your own scenario
Plan, execute, verify, and review before the work counts as done
A wiki your AI can use keeps sources, reviewed pages, and rules in files you own