There's a term developers have started using for what's piling up in their pull requests: AI slop.
Not an insult to the tools. A name for the specific kind of low quality, plausible looking code an agent produces when nobody's watching closely enough.
If you lead a small engineering team, you've probably felt this even without a word for it. Copilot, Cursor, Claude Code made writing code faster than ever. They didn't make the part that actually protects you faster: making sure what got written is correct, secure, and won't quietly become next quarter's incident report.
The numbers back up the feeling:
- AI generated code carries meaningfully more issues than human written code, including a higher rate of the critical and major kind
- One security study found AI coding tools producing vulnerable code in roughly 4 of 10 security critical tasks
- Review time on many teams now regularly beats writing time. You're not saving hours, you're just moving where they go
𝐓𝐞𝐥𝐥𝐢𝐧𝐠 𝐚 𝐭𝐞𝐚𝐦 𝐭𝐨 "𝐫𝐞𝐯𝐢𝐞𝐰 𝐦𝐨𝐫𝐞 𝐜𝐚𝐫𝐞𝐟𝐮𝐥𝐥𝐲" 𝐢𝐬𝐧'𝐭 𝐚 𝐬𝐲𝐬𝐭𝐞𝐦. 𝐈𝐭'𝐬 𝐚 𝐡𝐨𝐩𝐞.
The teams handling this well aren't reviewing harder, they're reviewing differently, routing effort to where AI is known to fail: logic edge cases, security patterns, architecture fit, long term debt. Not a generic checklist that treats all code the same.
I'm putting together a full framework for this: a 4 layer review stack sized for small teams, no enterprise tooling required.
Curious what's biting your team the hardest right now. Drop it below, it's shaping what I write next.
Top comments (0)