The No AI Slop manifesto from AIE Code Summit defines slop as output that costs the reader more effort than it cost to make
Engineers from the AIE Code Summit published a short manifesto on AI slop, and its central definition applies well beyond code.
What happened
The No AI Slop manifesto was written by speakers and attendees of the AIE Code Summit 2026 and introduced by Dex Horthy in the closing keynote, according to Latent Space's AINews. It models itself on the Agile Manifesto and lists five values, including outcomes over output, understanding over speed, and improving the system over patching the output.
Its definition is the useful part: slop is when someone puts less effort into producing something than they ask the consumer to spend understanding it. The authors say using AI doesn't by itself make work slop. It becomes slop when AI output is passed along without anyone taking responsibility for its quality.
The manifesto commits signers to a few concrete habits. When an agent wrote something being sent, they say so and check the parts that matter first. When a mistake shows up twice, they fix the system and its guardrails, and anything a linter, type checker or automated check can catch is left to that check. It also says agent reviews and verifiers help but are not enough on their own, so people should still review some subset of outputs based on the stakes. The site listed 352 signatures when I read it.
My take
In business automation, a lot of slop comes from workflows that generate text for people rather than from coding agents: a weekly AI summary of the pipeline that nobody reads, a follow-up email drafted from a CRM record with the wrong context, a ticket recap longer than the ticket.
The manifesto's definition gives a practical test for those workflows. If the recipient has to spend more time checking an AI written report than the old manual one took, the automation has moved work around instead of saving it. I'd fix that in the workflow itself: shorter outputs with a fixed structure, a link back to the source record so claims can be checked in one click, and a human review step for anything sent to a customer.
The point about fixing repeat mistakes in the system also maps directly to automation. When an AI step makes the same error twice, add a validation node or a schema check rather than tweaking the prompt again.
More posts
- Open OCR models now read 1,000 pages for under $0.10, and better serving saved more than switching modelsNov 20, 2026
- Gemini 4 Neon costs $3 in and $15 out per million tokens and tops Zapier's AutomationBench at 56.3%, though it uses more output tokens per taskNov 20, 2026
- Gemini API's Tool Retrieval loads tool schemas only when the model asks for them, so agents with hundreds of tools send fewer input tokensNov 20, 2026
