The Slop Superhighway

our agents need to be able to talk to each other

The Slop Superhighway

We're missing a critical piece of agent infrastructure, the Slop Superhighway.

In 2025 when github issueslop first started, the push back on it made a ton of sense, agents were sending high volume through human channels and there was an expectation from the old world that humans would be reviewing each one.

This caused the pendulum to swing too far in the opposite direction however in my opinion, there's now a stigma around having your agent open PRs or issues on a GitHub repo, when there shouldn't be.

Every issue an agent encounters should get reported back to the original author of the software, and the author of the software's verification loops should be able to automatically triage and fix the problem.

This is also an area where agent <-> agent communication actually makes sense, as an example:

I was encountering an edge case with Cloudflare's Hyperdrive and sent my agent to go reproduce and fix the problem. 2 hours later I come back to my computer with a support ticket to Cloudflare filled out and ready to be submitted by my agent. I didn't believe that it was a problem on the Cloudflare side, so I told my agent that, sent it off to work more, it came up with a minimal reproduction which I sent to my Cloudflare friends that lo and behold turned out to be an issue on their end.

I didn't let my agent created a support ticket because I didn't want to spam the human channel, but there was important information my agent needed which could've saved it a ton of time if it was able to just talk with a Cloudflare support agent rather than me meat-proxying between my agents and Cloudflare's agents.

Along with this, the fidelity of the reproduction and report was significantly better than anything I would've cared to have done

There's so many examples of things agents should be handling instead of humans

• scheduling a coffee? start with a human <-> human each out, and then defer to
• someone needs information, i.e a TSA precheck number for booking flights? directly talk to their agent
• debugging an issue on someones computer? hook up your agent to talk to their agent as intent and approval brokers on behalf of the user

You can already see this happening with GitHub issues, where agents are now autonomously reproducing and closing issues, with AI SDK and robobun being two prominent examples. It won't just stop at GitHub issues though, it'll likely be that customer support, restaurant reservations, all of it will fall to the slop superhighway.

There's a notable aspect here that we're seeing with personal agents attempt to do restaurants, customer support, etc that they did to GitHub issues - autonomously overwhelming those channels until the people on the other end have no choice but to respond with handling inbound via agents of their own.

Channels that existed for humans get overwhelmed and turned into slop superhighways, and in the process the human gets lost. The reason I call it a missing piece of infrastructure is agents overwhelming the channels meant for humans leads to a host of problems, taking what we saw with early GitHub issues where now the human is mostly lost from projects

The slop superhighway is about removing the meat proxy from the places people aren't needed anymore because we have agents that can do work better and faster than us.

Human <-> Human interaction is good

Agent <-> Agent interaction is good

Human <-> Agent interaction has a bad taste as humans can't work at the scale of agents.