People are generally amazing at finding use cases for technology that the originators hadn't thought of. I recall being surprised learning about Apache ZooKeeper being used for an online dating site, and similar situations have happened in several of the systems I had a chance to work on. The same seems to hold for the use of AI applications.
In an article from the Harvard Business Review publication (hbr.org/2026/06/how-peo…), I found out that “Astrology and Tarot Readings” is among the Top 10 use case categories of 2026 according to their study. Like any other nerd, I use it for coding tasks, processing logs, polishing writing maybe, but it would never occur to me that I could do astrology or anything of the like. Not that I'm into it, or have any kind of prejudice against it, it is simply not something I would immediately have thought of for AI. The top use case is therapy and companionship, which is more predictable given that there is a severe shortage of proper medical treatment in our society and solitude is a known issue especially at the later ages, although I'm skeptical about finding a good fraction of our elderly using AI for companionship now. Note that I'm not saying they shouldn't, only that there is a generational gap in the use of technology.
I learned a new term from the article, a concept that I observed but didn't have a single word for. "Thinkslop" is the "lazy, sloppy thinking that can be engendered by excessive use of AI". The study also refers to other relevant and often observed issues, like people stopping writing because AI does it, which impacts their reasoning. In fact, reasoning over some topic might deviate from its original intent as AI is doing the writing for you and even choosing the vocabulary. Or, the high-throughput generation of content that just goes to waste as no one actually consumes it. Of course, generating the content still consumes power, so the environment takes a hit.
But one that is a highlight for me is the false sense of intellectual rigor, which strangely reminds me of a Lego Movie quote that makes my kids and me laugh every time. When Vitruvius is talking about the prophecy, he closes it with "All this is true, because it rhymes", which is clearly absurd. The rhyme perhaps gives a false sense of correctness and rigor, which supposedly inspires credibility. How many times have you seen an AI report that looks fantastic but when you read it, it is mostly empty, a lot of words and maybe graphics saying very little and with slightly awkward vocabulary? It is not true only because it is AI.
Like many out there, I am making progress on what I consider a proper use of AI technology; it needs to be useful, an enabler even, but definitely constructive. It cannot work like a drug by giving you happiness out of a false sense of productivity. I am very sure that I will keep being surprised by use cases, and that's part of the fun.
I'd like to think that I get the point of shipping fast rather than getting stuck on making it perfect, but what's the point of shipping something that can make your customer lose money, critical data or perhaps even cause an accident? If we are talking about "the rightmost button is red rather than blue" kind of broken, then I can understand the observation, but it is not generalizable, is it right?
I don't know much about the LoC use case, but my initial reaction to the post is to ask why they are not building a data lake with open formats. I'm sure there are reasons for discarding open-table formats. Claude keeps telling me that the issue is that they don't address preservation properly.
Thanks for catching the point on AI-native access. Previously, operating a Qbeast-backed lakehouse required terminal commands; we've now added MCP connectivity to enable agentic coding tools, see the linked post for details and screenshots. Query optimization comes via Qbeast's indexing and auto-optimize features. Semantic layers aren't on our radar yet, but we're open to adding them to the roadmap. Keep in mind we're an early-stage startup, so we're just getting going.
Qbeast is an efficient, format-agnostic, multi-dimensional clustering engine for the lakehouse, now delivered as a distributed service. Our latest release marks a major milestone: Qbeast runs on all three major cloud providers (AWS, Google Cloud, and Azure) and supports all three open table formats (Apache Iceberg, Delta Lake, and Apache Hudi), with fully automated deployment and management inside the customer's account. The release brings substantial performance gains in both querying and indexing, and introduces a Model Context Protocol (MCP) server that enables direct integration with AI tools—making Qbeast a first-class data source for agentic workflows.