Plenty of the things you want to know about a prospect are not on their website. Funding, recent hires, which conferences they exhibit at, whether they just opened a second location — those live in press coverage and directories.
The research agent now has web search available as a tool alongside site crawling, so it can go and look. It also handles the messy parts of the open web better: sites that block datacentre traffic are fetched through a proxy, and failures now surface a real error on the run log instead of an empty field.
Fewer blank answers, and the ones that do come back blank tell you why.