Blog — SwiftAct

October 9, 2026

Research A context analysis of a failed code-review revision: Kimi K3 and LLVM PR #220979

Truncating the original session at the decision point and resampling with single-variable edits: without history K3 applied the review 5/5, resuming unedited 1/5, and one explicit sentence of review intent restored it to 3/3.

Read ↗

new ★

June 5, 2026

Opinion Why smarter Agents still lead to terrible coding

AI agents speed up software development, but push systems into uncontrollable and unmaintainable chaos even faster.

Read ↗

May 26, 2026

Engineering Harness Engineering: Leveraging Kimi Code in an Agent-First World

Starting from kimi-code-swarm, this article reviews intrinsic failure modes in multi-agent coding and how mechanical constraints and gating reduce technical debt.

Read ↗