A quarterly read of the catalog and the conversation around it: what evaluators are searching for, what makers are shipping, and where the money is moving. The interesting signals this quarter are structural, not volume-based.
Consolidation, not explosion
The tool count keeps rising; the attention is consolidating. In almost every category the top five listings now absorb the majority of outbound clicks, and the long tail is being priced out of relevance rather than out of existence. The "model wrapper with a pricing page" tier is disappearing: evaluators have learned to spot it in one screenshot, and the fifteen-minute evaluation checklist is doing to thin tools what review aggregates did to thin apps.
Agent frameworks grew up
Automation & Workflow Orchestration has quietly become the ecosystem's center of gravity. The tools that survived the year share a shape: durable state, human checkpoints, and honest failure modes. The conversation moved from "what can an agent do" to "what does an agent do when it fails", which is what maturity sounds like.
Local models went mainstream
A year ago local inference was a hobbyist flex; now the Developer Frameworks & Infrastructure category treats it as table stakes. Privacy-sensitive buyers (legal, healthcare, finance) are the driver: the pitch "your data never leaves the building" now closes deals that cloud-only vendors used to win. Expect the local-versus-API tradeoff to be a standard comparison filter within the year.
Voice cloning hits its compliance era
Voice moved from demo to deployment, and immediately collided with consent law. Watermarking, per-voice consent records, and cloning opt-out registries are becoming procurement requirements, not differentiators. The tools that treated this as a product feature early are the ones showing up on enterprise shortlists now.
One number to remember
Seven categories. We rebuilt the directory's taxonomy this month around the seven that describe how evaluators actually shop for AI: conversational, generative content, NLP, vision, data, automation, and developer infrastructure. When the shelves match how people think, discovery gets boring in the best way: you search, you compare, you get back to work.