"Let's dive in." "I'd be happy to help with that." "While I can't provide real-time information, I can tell you that..."
We all know the cadence. The GPT-isms. The passive corporate helpfulness that every model defaults to. It doesn't just read as robotic anymore — it reads as lazy interface design. And today, two independent projects hit the same nerve simultaneously: users are done with AI writing like AI.
claudish-to-english hit GitHub trending at #3 with 579 stars — a Claude Code plugin that rewrites every assistant message into plain English using a local ollama model. Not a prompt hack. Not a system message. A display-layer filter that strips the "I'd be happy to" before you see it. And on Hacker News, Claudette (163 points) — a project with the tagline "Make Claude stop talking like a BuzzFeed article."
These aren't competing. They're converging on the same insight: the default AI voice is costing attention.
This isn't about politeness. It's about signal-to-noise ratio. Every "certainly!", every "great question!", every "I appreciate your patience" is a token spent on nothing. When you're asking a code agent to debug a segfault, you don't want warmth. You want the fix.
The timing with the Economist study (196 HN points) about AI-boosted homework scores but dropped exam scores isn't coincidental. The same dynamic carries over to professional work: AI makes you faster at the surface level while your ability to produce clean, direct output — without it — atrophies.
The anti-AI-ese movement is still early, but the signal is real. When the tooling to strip AI voice is trending on GitHub and HN on the same day, the market is voting. The next wave of AI products won't compete on capability alone. They'll compete on how little they sound like AI.