> “Disorganized data silos” have been an issue for Copilot, analysts wrote.
This is true in almost every large organization, and will affect every enterprise AI product out there. There was a relevant subthread just a couple of days ago recounting this exact, same dynamic: >>46861209
In fact, Palantir's secret sauce may not be their tech, but their "Forward Deployed Engineers" model (i.e. a rebranding of "partner engineers embedded within their customers' organizations"). Because it turns out that's a lot of what they do is navigating these bureaucratic and political hurdles to unlock access to the data: https://nabeelqu.co/reflections-on-palantir
It gets even worse if you consider this data is going to be extremely messy, with multiple bespoke, partially-duplicate / overlapping, potentially conflicting versions of the data with varying levels of out-of-datedness, scattered across these silos. (I would know, in a past life, I worked on a months-long project called, self-explanatorily enough, "Stale Docs".)
Yeah, untangling these bureaucratic webs and data horrors is not a quarter-long or year-long project, so investors are gonna be waiting a long time for the impact of AI to be visible. On the bright side, as TFA also hints at, AI providers themselves have been severely capacity-constrained. So hopefully by the time these issues get sorted out enough new capacity would be coming online to actually serve that traffic.
In the meantime, I expect a prolonged period of AI companies feverishly splurging on AI CapEx spend even as Wall Street punishes them repeatedly for the lack of impact of AI being reflected anywhere.