When looking at business research and the use of agentic AI, there is, in my opinion, a parallel between the emergence of agentic AI and the early days of digitizing paper-based resources such as newspapers, journals and card-based library catalogs. There was no doubt with information professionals who understood in those early days the benefits of digitalization and that it was the future, regardless of any misgivings raised. One critic, Nicholson Baker, who famously wrote a controversial essay lamenting the demise of card catalogs, believed they housed important information not captured in their “electronic surrogates” (basically the stripping away of context). He stated, “We’re so infatuated with this new way of storing things that we have momentarily forgotten ourselves.” This is a sentiment that is still ongoing about not getting blinded by the buzzy appeal of shiny new technology.
Now, as we look at agentic AI for business research, we understand that this is the future, even with misgivings. For an example of how this looks now and in the future, a McKinsey case study features a research firm’s use of an agentic approach. The solution uses autonomous multiple agents that identify shifts in sales or market share by analyzing internal factors and external events identified via web searches. The solution then synthesizes and ranks the most influential drivers. Advanced search and contextual reasoning capabilities can uncover insights that are difficult for human analysts to find manually (text is summarized, not quoted).
There is a clear distinction between AI agents and AI agentic solutions that are built using data from sources where curation is done with authority in mind, usually from commercial or fee based databases and repositories. LLMs that have been trained on domain specific data are more likely to produce better, higher quality results, especially for areas such as law, medicine, and business/finance. These solutions are already starting with a solid foundation that ensures data quality, or as I refer to as a strong source foundation.
However, concerns abound for sources that are scraped from the internet, where there is little authority control. In terms of business research we are forgetting that Ai agents and workflows built on shaky foundations due to poor data quality will lead to building entire houses of cards that ultimately will collapse and be made useless. Christophe Lederrey, from Aetheris, on his LinkedIn feed, best sums up why so many companies struggle to operationalize AI. “They believe AI will fix their data issues — but in reality, it magnifies them…Fix the foundations first. Then AI becomes trustworthy…”
The good news is that there is innovation addressing source credibility. However, building AI agents that verify and fact check source information is not enough. The most accurate business research results are those that come from sources that best meet and match the research need at the time, thus the best resources must be continuously made available. What is needed are AI agents and workflows that are designed to find the best and most relevant research sources that match the immediate need at the time. What good is fact checking if its being done on less relevant/qualified sources that do not meet the actual need? This can be quite challenging for many reasons. Two of which are:
- It is unclear exactly which sources are, and are not, available for inclusion in an LLM model. Website publishers can block web crawlers from accessing their content and many credible ones do not provide permission, nor license their content for LLM training purposes. Publishers are now ramping up efforts to protect their websites “from tech companies that hoover up content for new AI tools.” Keeping up with developments in this area is overwhelming.
- Unfortunately, fake news sources are here to stay. Just this month, more than 4,000 fake gen AI powered websites (mostly in French but expanding in English) were built to “spew AI-written articles plagiarized from other sites or which are simply made up with the aim to either earn money by ‘backlinking’ (artificially boosting the rankings of other sites) or by appearing in Google Discover to harvest ad revenue.”
In terms of business research, I believe that agentic AI can and will move us forward. What is required are solution offerings that are built and based on solid source foundations.
Photo by Mohamed Nohassi on Unsplash
July 17, 2025
