FIELD NOTES
Notes from the lab
Measured findings from building AI agents and cloud architecture: what we tested, what it cost, and what we would do again.
Compaction and the prompt cache
What compaction does to an agent's memory and its prompt-cache bill, how Microsoft Agent Framework's built-in strategies hold up on a retrieval-heavy conversation, and the benchmark we built to measure it.
Read the article →Compaction that keeps the cache
The compaction strategies we built for Microsoft Agent Framework: the rules behind them, how the composed strategy works, where each one wins, and which to use for which case.
Read the article →Using the compaction strategies
A hands-on guide to the strategies from Part 2: what each one does to the conversation, the settings that matter, the counters to read afterwards, and the code to attach each one to a Microsoft Agent Framework agent.
Read the article →