A business analyst creates a custom agent to draft competitor summaries. Drafts must cite current pricing from vendor sites, but the agent currently uses only training data. Which change should the analyst make to create a custom agent by adding web search to instructions or prompts?

The correct response succeeds because it directly remedies the stated gap by embedding web search into the standing instructions and pairing live retrieval with an in-text citation obligation for every output, which ensures pricing reflects vendor sites at generation time rather than static training data. The weekly-cached variant fails on currency because scheduled snapshots preserve consistency and reduce tool calls, a sound efficiency pattern, but stale prices cannot satisfy a current-pricing mandate when vendors change plans mid-cycle. The separate-audit-log variant fails on placement because centralized source tracking aids governance and is widely used for reviewability, yet a citation stored outside the deliverable does not meet the requirement that each summary itself substantiate its numbers for the reader. The aggregator-site variant fails on authority because comparison hubs are efficient for discovery and are commonly used for market scans, but intermediary listings can lag, misinterpret tiers, or omit regional terms, so they cannot substitute for primary vendor pages when vendor provenance is required. As an exam caveat, distinguish frequency, placement, and provenance as independent constraints rather than treating any web-enabled workflow as sufficient, since prompts often make all choices sound retrieval-aware while only one satisfies all three together. As a practical check, pilot the revised agent on two competitors with a recent price change and confirm each draft contains a clickable vendor-domain citation adjacent to the price that resolves to the live figure before wider rollout.