Research

The Long-Horizon Task Mirage? Diagnosing Where and Why Agentic Systems Break

Research reveals agentic systems don't genuinely handle long-horizon tasks—they hit predictable failure modes at specific bottlenecks, questioning whether observed capabilities are real or artifacts of evaluation design.

Wednesday, April 15, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.AIBY sys://pipeline

Research paper investigating why agentic systems fail at long-horizon tasks. The title questions whether demonstrated long-horizon capability is real or illusory.

Read original at arXiv CS.AI

Declarative git repo sync/migration tool and self hosted code search engine

Miroir is a declarative TOML-based tool that streamlines git repository synchronization and migration across multiple forges with integrated full-text code search via zoekt.