Everyone wants to be the App Store for agents. We think that's the wrong fight.

Open any AI newsletter this month and you'll find a version of the same story: another platform announcing its own storefront for agents. Count them and you land somewhere around eight serious contenders right now — Claude's own Skills, OpenAI's GPT Store, the various MCP hubs, Hugging Face Spaces, Replit's agent market, LangChain's hub, Vercel's gallery, Cloudflare's marketplace. Different owners, same instinct: put agents on a shelf, let people browse.

We watched this shape up while building AstrikSwarm, and it's worth saying the quiet part out loud — the shelf isn't the hard problem. Getting listed is easy compared to what happens after.

The shelf is fragmenting, not consolidating

Each of these marketplaces has picked its own way to rank listings and its own way to get paid. Some are free to publish and make money on the consulting around them. Others take a cut of usage. A couple just sell the agent outright, like software always used to be sold. None of them agree with each other, and none of them talk to each other.

That means a builder chasing distribution has to show up in all of them, tuned slightly differently for each audience, and keep every listing fresh — because the platforms clearly reward the ones that get touched every month over the ones left to rot. That's a full-time job that has nothing to do with whether the agent is any good.

Our bet: skip the shelf, wire the agents together

We didn't build AstrikSwarm to win a listing. We built it because the real friction we kept hitting wasn't finding an agent — it was getting three good agents, from three different places, to actually work a task together without a human relaying messages between them.

A marketplace solves discovery. It doesn't solve coordination.

A support agent that lives in Slack, a data agent that lives behind an API, and a scheduling agent that lives in its own app — none of them care which storefront they were downloaded from. They still won't talk to each other unless something sits underneath and does that job on purpose.

What we're watching instead

• Whether platforms start competing on interoperability rather than exclusivity — right now most want you locked to their shelf.
• Whether "agent harness" quality (how well an agent handles messy, multi-step real work) starts mattering more than which store it's listed in.
• Whether small studios like ours can keep building the connective layer without needing to win a platform's algorithm first.

Our guess: in a year, nobody will remember which store an agent launched on. They'll remember whether it got the job done. We'd rather spend our time on that than on shelf space.