I work in product at a Series B and we cancelled most of our AI subscriptions this quarter

We bought everything when the hype was at its loudest, ChatGPT enterprise for the team, Claude through the Anthropic API for the eng side, Notion AI, Mintlify for the docs, Cursor for the engineers, BuildBetter for customer feedback, Otter for meeting notes, Perplexity for research... 8 line items on the company card and none of them felt optional in the moment we clicked subscribe. we pulled the spend and cancelled the ones the team had stopped opening, and ChatGPT survived and so did Cursor, and there was one fight we lost with the CX team over a smaller customer feedback tool they refused to give up, and everything else is gone. I'm not sure if we were idiots for buying all of it or if the AI category is just structurally bloated right now (probably both) and the thing that's hard to say out loud is that most of what we tried did basically the same job as ChatGPT or Claude with a thinner wrapper on top. The ones that survived are the ones that do something the foundation models don't. The honest take I keep handing junior PMs at smaller companies is that if a vendor is pitching you AI tooling, ask what you would lose by just using a foundation model directly. If the answer is fuzzy the tool will be on the next cut list.