Maligned #10 - The Hype Cycle Turns
Monday. New week. Let’s talk about where things actually stand.
We’re entering the “now what” phase of AI
The initial wave of excitement has crested. Every company has an AI strategy deck. Most of them have at least one pilot project running. And a lot of those pilots are stuck. The pattern I’m seeing repeatedly: the demo worked, the POC looked promising, and then the team hit the wall of production requirements. Data quality, latency constraints, cost at scale, edge cases, monitoring, security review. This is the boring, essential work that separates real AI deployments from expensive science projects. The companies that invested in their data infrastructure before the AI hype are reaping the benefits now. Everyone else is trying to backfill, and that takes time.
The consolidation of AI labs is accelerating
The number of organizations that can credibly train frontier models is shrinking, not growing. The capital requirements are enormous and increasing with each generation. We’re settling into a pattern with maybe five or six labs at the frontier, a larger tier of companies fine-tuning and adapting those models, and a massive long tail of application builders. This isn’t necessarily bad, but it does concentrate a lot of power in a small number of organizations. If you care about the long-term structure of the AI industry, this consolidation trend is the most important thing to watch.
Retrieval-augmented generation is table stakes now
RAG went from novel technique to standard practice faster than almost any other AI pattern I’ve seen. Every enterprise deployment I look at uses some form of it. The tooling has matured enormously. Vector databases are stable and performant, chunking strategies are well-understood, and the reranking and filtering layers that used to require custom engineering are now available off the shelf. The next frontier is making RAG systems that actually know when they don’t have enough context to answer well, rather than confidently generating plausible nonsense from irrelevant retrieved documents.
A note on this newsletter
This is issue 10 of Maligned. I started this because I was tired of reading AI coverage that was either breathless hype or doom-and-gloom fear. The reality is more interesting and more mundane than either extreme. AI is a powerful set of tools that will change a lot of industries, slowly and unevenly, with plenty of false starts and dead ends along the way. If that kind of clear-eyed perspective is useful to you, stick around.
See you next week.
Maligned - AI news by Mal