Agents

Published on
June 13, 2026
I tried to disprove the TDD benchmark. I reproduced it.
agents benchmarks tdd codex programbench
I expected a better TDD skill to beat Kun Chen's ProgramBench result. It did not. A behavior-first TDD variant tied the original TDD arm and still trailed the no-TDD control.
Published on
May 6, 2026
Calling it memory made agents better at memory
agents memory longmemeval benchmarks napkin
Napkin was already strong on LongMemEval_S. Renaming the model-facing surface from napkin to memory improved one Qwen3.6 run from 87.6% to 90.6% accuracy, with the biggest gains on multi-session and temporal reasoning.

I tried to disprove the TDD benchmark. I reproduced it.