Spotify has stopped pitching its algorithm. That’s the quiet shift buried in this week’s coverage. For years the company sold you on Discover Weekly, on the idea that its recommendation engine knew your taste better than you did. Now the pitch has moved. Spotify wants to be the music service that an AI agent picks when you say “play something for my commute.”
That’s a different fight entirely. And it’s one every app maker should be watching.
The click is dying, and agents are the reason
The old app economy ran on a simple loop. You searched, you scrolled a list, you tapped an icon. Developers paid for placement, optimised store pages, and begged for reviews. Every step assumed a human was making the final call.
Agents break that loop. When you tell an assistant to book a table, queue a playlist or reorder printer ink, you rarely see the shortlist. The software chooses, and you get the result. Spotify’s agentic voice push is built on exactly this logic, with the idea that voice is the new click.
It’s already happening in small ways. Spotify’s ChatGPT integration went live in 145 countries, which means the service now lives inside someone else’s interface. Anthropic’s Claude has added Spotify and StubHub as connectors too. Streaming rivals aren’t just competing with each other anymore. They’re competing for a slot in the assistant’s default behaviour.
Here’s the uncomfortable part. A default is nearly invisible. If the agent reaches for Spotify, Apple Music never gets a look, however good its catalogue is.
What an agent actually weighs
So what does a software agent use to pick a winner? Nobody outside the labs knows the exact recipe, but the likely ingredients are easy to guess.
First, whether it can actually do the task. Does the app expose a clean connector, a reliable API, a permission flow that doesn’t break? Those are the software and apps basics that decide whether an agent can use a product at all. Second, user history. If you’ve played Spotify for six years, the agent leans on that. Third, and this is the one that matters for everyone who isn’t Spotify, trust signals pulled from the open web.
Agents don’t have taste. They borrow it. When a user asks a question where the answer isn’t obvious, the assistant tends to lean on sources that look authoritative, are cited by others, and carry a named human or a regulated reputation behind them. Raw rankings and keyword-stuffed listicles are the cheapest signals to fake, so a sensible agent discounts them.
That has consequences well beyond music. It reshapes who wins in any category where the stakes are real: money, health, legal rules, anything where a bad recommendation costs the user something.
A test case: regulated products where trust is the whole product
Take a category that is heavily regulated and awkward to recommend responsibly. Online blackjack in Australia is a decent example, because the legal picture is genuinely messy, the operator quality varies wildly, and a careless answer from an agent could point someone at an unlicensed site.
An assistant asked “where can I play blackjack online in Australia?” faces a real problem. Scrape the top ten search results and you’ll mostly get affiliate pages optimised for clicks. A careful agent, or a careful developer building one, would prefer a source with editorial accountability. A masthead. Named reviewers. A track record of covering the local market rather than recycling a template.
That’s the sort of source a well-built agent would likely cite, and Stevivor’s online blackjack guide fits the profile: an established Australian outlet publishing a dedicated guide to the topic, rather than a throwaway ranking page. Whether any given agent actually cites it today is a separate question. The point is the pattern. Editorial guides with a real publisher behind them are the kind of thing agents are likely to prefer over anonymous ranking pages.
Adults only, and a quick reminder: gambling involves risk, so play responsibly and only with money you can afford to lose.
The squeeze on mid-tier apps
Now back to the software itself, because the picture gets grim for anyone who isn’t a household name.
If agents default to the incumbent, the mid-tier app has two bad options. It can pay for a connector slot and hope the assistant surfaces it, or it can try to win on a narrow task the big player handles poorly. Small-business software vendors know this feeling already from marketplace fees. Agents just move the toll booth somewhere you can’t see it.
Spotify’s own position isn’t comfortable either. A TechCrunch analysis argued that its AI bet risks delivering more of everything and less of what listeners actually want. That tension matters. If the agent layer optimises for engagement rather than satisfaction, the app being chosen and the app being best can drift apart.
There’s a second risk. An agent that always picks the same service teaches users nothing about alternatives, which flattens the market. Great for the incumbent, miserable for everyone building something stranger and better.
What builders and publishers should do now
If you make an app, treat agent compatibility as a product feature, not a marketing checkbox. Ship clean connectors. Keep your documentation machine-readable. Make it dead simple for an assistant to complete a task end to end, because the agent will take the path of least friction.
If you publish guides and reviews, the advice is almost old-fashioned. Put a real name on the work. Show your testing. Cover a region or a niche deeply rather than everything shallowly. Agents are being built to avoid slop, and Spotify’s own messaging is explicitly about preventing it. That filter will cut against thin content first.
And if you’re just a user, ask your assistant why it chose what it chose. Most will tell you. The answer is often more revealing than the recommendation.
The Spotify bet in plain terms
CNBC framed the Spotify and ChatGPT tie-up as a competitive bet against Apple, Amazon and YouTube, and as part of OpenAI’s push to turn ChatGPT into a platform. That’s the cleanest reading. Spotify isn’t betting on AI features. It’s betting on being first in line when AI does the choosing.
Will it work? Probably for Spotify. For the thousand smaller apps downstream, the lesson is harder. Visibility is no longer about being found. It’s about being picked, by something that never sees your logo.
FAQ
What does it mean for an app to be “chosen by an AI agent”? It means the assistant, not the user, selects which app completes a task. You say “play something relaxing,” and the agent decides between services. The winning app often never appears on screen as an option, which makes defaults and integrations far more valuable than store rankings.
Why is Spotify pushing agents instead of its recommendation algorithm? Because people increasingly talk to assistants rather than browse apps. Spotify’s pitch has moved from “our algorithm knows you” to “our service is the one agents reach for.” Being embedded in tools like ChatGPT and Claude secures that position before rivals do.
How can smaller apps compete in an agent-driven market? Build reliable connectors, document everything clearly, and own a narrow task better than the incumbent does. Agents favour options that work end to end without friction. Specialists with clean integrations can win niches that big platforms handle in a generic, clumsy way.
Will AI agents trust review sites over rankings? Likely, where stakes are high. Agents are being tuned to avoid low-quality content, so sources with named authors, real testing and an established publisher look safer to cite. Anonymous, keyword-driven listicles are the easiest signals to fake, so they tend to get discounted.
Should publishers change how they write for agents? Mostly by doing the basics well. Show who wrote it, show the testing, and cover a topic with real depth. Structure helps, but substance matters more. Agents borrow taste and trust from the open web, so editorial credibility is the durable advantage.
The next few years will reward whoever agents trust, not whoever shouts loudest. Spotify understood that first. Everyone else is catching up.