Sloppypaste has a clean pitch: paste this URL to anyone who's ever forwarded you a 700-word Claude response with no context and the implicit instruction to figure out which part of it matters. The awareness site arrived this week with a deliberately minimal footprint — a term, a definition, a tagline. 'Read it before you share it. Don't be a slop cannon.' The tone is gentle enough that forwarding it to an offending colleague skips the awkward conversation.
The name lands. The behaviour is real and common enough that most knowledge workers who've spent time in an LLM-heavy team will recognise it instantly — an unedited wall of model output dropped into a Slack thread or email chain, with no indication of what the sender actually wanted you to do with it. But sloppypaste.com isn't purely a cultural intervention. It's also a product landing page, and the product is Agent Relay: a real-time communication layer designed to let AI agents pass information directly to each other, cutting out the human copy-paste step entirely.
The reframe is the interesting move. Rather than treating sloppypaste as a behaviour to correct through better etiquette, the site positions it as a symptom of broken workflow architecture — if agents could relay outputs autonomously, the problem disappears at the source. Whether that framing reflects a real technical solution or mostly serves to make the product sound inevitable is harder to judge from the outside. The Agent Relay section is vague on specifics: no integration documentation, no indication of which agentic stacks it supports, no pricing, no case studies or named customers. Requests for comment sent to the contact address on the site were not returned before publication.
What would strengthen the case considerably is data. How widespread is the sloppypaste habit, and is it a minor annoyance or a measurable drag on team output? The assertion that informal norms around AI-generated content are only now beginning to form may well be accurate — it broadly matches what's observable across the industry — but asserted without evidence it reads like scene-setting rather than reporting. A few figures, or even anecdotes from named practitioners, would do a lot of work here.
None of that necessarily undermines the neologism itself, which has a reasonable shot at sticking regardless of whether Agent Relay ships in any meaningful form. Naming a friction point is often how software categories get started: someone articulates the pain clearly enough that others recognise it, and the product follows. The agent ecosystem is still largely running on improvised human relay steps, and there's genuine space for infrastructure that makes inter-agent handoffs less dependent on someone manually shuttling transcripts. Whether that infrastructure turns out to be Agent Relay is a different question.