Context
An original research preprint introduces Programmatic Search Agent (PSA), enabling agents to retain candidate evidence and compose retrieval, filtering, reranking, extraction and selective presentation operations. Across two deep-search benchmarks and five model backbones, PSA improved task success over a query-based search interface in a controlled evaluation.
What changed
The study isolates an evidence-delivery bottleneck: relevant supporting passages can be retrieved but omitted from an agent's observations. PSA allows executable processing of retained candidate pools and selective evidence presentation, rather than relying only on successive query reformulations.
Why it matters
For AI discovery, a page entering retrieval is not equivalent to its supporting passage reaching the model or being used in an answer. The research supports separating retrieval, evidence delivery and final source use in visibility and attribution measurement.
Visibility OS interpretation
Visibility OS interpretation: research into agent-controlled evidence processing broadens the query-fan-out narrative; discoverability is partly determined by which evidence survives post-retrieval processing. This is a mechanism hypothesis for publisher visibility, not a confirmed commercial search ranking signal.
What remains uncertain
Will production AI search engines expose persistent candidate workspaces or selective evidence presentation? Do these controlled benchmark gains transfer to live, dynamic web corpora and citation allocation? How do total end-to-end compute and costs compare?