Gisting agent context for higher throughput
A learned representation cuts serving work while retaining task-relevant state.
AI systems
Compress agent context. Keep useful work.
A learned token layer preserves useful context while cutting the work passed to the serving model.
Latest writing
Raising mobile end-to-end test stability
How better fixtures, ownership, and failure grouping made the test signal useful again.
Keeping queue retries predictable
A review of idempotency, visibility timeouts, and observable failure handling.