The following slides I presented on PyCon 2026 as lightning talk.
Vimeo Link - jump to 05:45:00 to watch the presentation.
Stop the Slop!
A 5-Minute Call for Action for the AI-Assisted Workplace
PyCon DE 2026 — Lightning Talk
It's Official.
"AI slop": Merriam-Webster 2025 Word of the Year.
Low-quality digital content produced, usually in quantity, by means of AI.
Like "spam", but now it's coming from your team.
Your colleagues, PRs, tickets, chat.
No spam filter, no mute button, no protection.
Nobody is safe.
The Numbers
- 40% of workers received "AI slop" in the past month
- 15% of all content received at work is AI slop
- $9M / year in lost productivity for a 10k-person org
- 61% hide their AI use from colleagues
- 55% pass off AI output as their own work
- 2/3 never check AI output before sending it
-> AI should save time. Now, we're drowning.
curl Rage-Quit
- Daniel Stenberg (curl maintainer) shut down a bug bounty after too much AI slop:
"We now ban every reporter INSTANTLY who submits AI slop. We are effectively being DDoSed."
- curl's
security.txtnow reads:
"We will ban you and ridicule you in public if you waste our time on crap reports."
- 📊 20 submissions in 16 hours. None were real vulnerabilities.
Code Reviews Gone Wrong
- A study of 1,1k developer posts:
Individual devs ship faster. Reviewers, maintainers & the community pay the price.
- Example: An AI agent hallucinated external services, then mocked out these services, i.e. an internally consistent but verbose hallucination.
"They're literally just using you to do their job — critically evaluate their AI slop and give it the next prompt."
What Can You Actually Do?
As a colleague:
-
Be transparent — "I used AI for this" lets collaborators calibrate effort.
- BTW: I actually used AI for this presentation
- Keep generations small — Every slide in this presentation is short and thus easily quality assured.
- Manually refine — Read all AI output. Make sure you understand it and remove unnecessary and wrong information.
What Can You Actually Do?
As a developer:
- Two-prompt rule — If the 2nd attempt doesn't improve, stop prompting. Write it yourself.
- Keep generations small — If you can't explain the output, reject it.
- Be transparent — "I used AI for this" lets reviewers calibrate effort.
- Don't dump AI plans on teammates — Pasting a raw Claude plan into a ticket is asking others to do your work.
What Can You Actually Do?
As a team:
- Review = understanding — If no one understands the code, it doesn't ship.
- Name it — If you see slop, say "this looks like slop." Normalize the feedback.
- Rule of thumb: If you can't defend it in a review, don't submit it.
- Establish best practices: DSPI (look it up!)
Remember:
AI slop is not a technology problem.
It's a laziness problem.
The tool is fine. The copy-paste is the crime.
Resources:
- 🌐 stoptheslop.dev — Internal dev team policy template
- 📄 arxiv.org/html/2603.27249 — "An Endless Stream of AI Slop" (academic study)
- 📝 daniel.haxx.se/blog — curl maintainer's AI slop chronicles
- 🔧 github.com/hardikpandya/stop-slop — Claude skill to de-slop your prose
Sources:
- BetterUp Labs / Stanford Social Media Lab (2025) — hbr.org/2026/01/why-people-create-ai-workslop-and-how-to-stop-it
- Global survey of 32,352 workers across 47 countries — theconversation.com/ai-workslop-is-creating-unnecessary-extra-work-heres-how-we-can-stop-it-267110
- Heidelberg / Melbourne / Singapore, 2026
