September 23, 2026
YouTube announced at its Made On YouTube event that users will soon be able to describe, in plain language, what kind of feed they want, and Gemini will build it. Spotify moved in the same direction the same day, launching Taste Profile in the U.S. -- a feature that shows Premium subscribers how the app thinks it understands their taste and lets them rewrite that understanding using natural language. YouTube also rolled out 'Ask Music' and 'Your Podcast Lineup' for conversational discovery, and gave creators in YouTube Studio new AI tools for generating thumbnail ideas and tracking how they perform.
For years, the pitch from every recommendation engine was 'trust us, we know you better than you know yourself.' That pitch is quietly being retired. Both companies are conceding that opaque algorithms frustrate as often as they delight, and that giving users a visible, editable model of their own preferences builds more trust than another layer of invisible optimization. That's a meaningful shift, not a cosmetic one. It also happens on the same day these two platforms -- longtime rivals for listening time -- ship almost identical ideas, which tells you this isn't one company's clever feature, it's where the whole category is heading.
The business lesson generalizes past music and video. Any product that makes decisions on a user's behalf -- a CRM scoring leads, a support tool triaging tickets, a dashboard prioritizing tasks -- is going to face the same pressure to become conversational and steerable rather than a black box. Teams building internal tools should take note: an AI app builder that lets non-technical staff describe what they want a workflow to prioritize, in their own words, is doing for internal operations what YouTube just did for a video feed. The interface pattern -- describe your intent, let the model configure the system -- is becoming the default expectation, not a novelty.
While consumer apps get the headlines, TechCrunch reported that enterprise AI agent startup Ema raised $77 million, bringing its total funding to $140 million, with more than 50 enterprise customers including Google and Microsoft. The framing matters: this is being described as AI starting to eat into enterprise software and services spending, not just augment it. That's a bigger claim than another funding round usually deserves, and it lines up with something we've been tracking here -- vendors like Salesforce making the case that off-the-shelf agent products can replace swaths of what used to require custom software or outside services firms.
I think this is the more consequential story of the day, even though it won't trend as hard as anything with 'YouTube' in the headline. Consumer recommendation tweaks are a UX improvement. Enterprise AI agents landing at companies the size of Google and Microsoft, displacing budget that used to go to software licenses and services contracts, is a market restructuring. It's the same dynamic we've discussed with replacing bloated per-seat software stacks -- once an AI layer can do the configuring, integrating, and reporting that used to require buying five separate tools or hiring a systems integrator, the pressure on legacy per-seat pricing only grows. If you want the deeper math on why that pricing model is getting harder to defend, we laid it out in our recent piece on per-seat SaaS costs.
The honest counterpoint: $77 million is a healthy raise, not proof of a category-wide collapse, and enterprise software incumbents have survived plenty of 'this changes everything' cycles before by acquiring or copying the threat. Ema still has to prove retention and expansion at those big logos, not just landing them. But the direction of travel -- from rigid, seat-priced software toward AI that configures itself around what a business actually asks for -- is the same direction YouTube and Spotify just moved their own products. Control is shifting from the platform to the person typing the request, whether that person is a listener or a procurement lead.
If your team could rewrite how your own tools prioritize work just by describing it in plain language, the way YouTube and Spotify users now can with their feeds, what's the first workflow you'd want to fix?
Sources