AI agents can falter on longer runs, a post summarizing a Microsoft paper says

A post summarizing a Microsoft paper says success usually declined across nine models as the number of dependent steps increased, with small errors compounding over longer workflows. It says shortening context made the decline worse, so trimming history was not a reliability fix. The user recommends testing agents at realistic workflow lengths, measuring per-step reliability and adding checks or checkpoints—rather than treating benchmark pass rates as proof of production readiness.