|
|
Was this forwarded to you? Get The Fourth Estate AI Brief in your inbox →
|
|
|
|
View this issue in your browser
Tuesday Brief
The 75% Question: What Google's AI Code Numbers Mean for Your Newsroom
The interesting thing this week with AI and publishing — Google announced that 75% of its code is now AI-generated, up from 25% just eighteen months ago. That's not a typo. This trajectory tells us something important about where the human-plus-AI workflow is heading... and how fast it's getting there.
|
Google's 75% — and Why the 'Still Reviewed by Humans' Part Matters Most
Google dropped a number this week that caught my attention: 75% of their code is now AI-generated. But here's the part that matters for us — every line still gets reviewed and edited by human engineers. They're not replacing programmers. They're changing what programmers spend their time doing. Sound familiar? At my paper, I don't want AI writing obituaries from scratch. I want it handling the formatting grunt work so our staff can actually talk to families and get the details right. Google's model is basically what we should be aiming for: AI does the heavy lifting on repetitive tasks, humans do the judgment calls and quality control.
When skeptical staff ask 'are you trying to replace us with AI?' — point to Google's 75% number and explain that their engineers still review everything; the byline still means something.
Source →
|
Which AI Tool Should You Actually Use? A Framework That Makes Sense
I get this question constantly from other publishers. 'Should I be paying for ChatGPT or Claude or... what's that Perplexity thing?' Here's a framework that finally made sense to me: Claude for longer writing projects and editing (think grant applications, feature rewrites, those 2,000-word anniversary pieces). ChatGPT for quick brainstorms and everyday questions. Perplexity when you need to find sources fast — like when you're backgrounding a city council agenda item at 4pm. And NotebookLM if you want to interrogate your own archive. The key advice? Start with one. Don't subscribe to four services and use none of them well.
The best AI setup isn't the most comprehensive one — it's the one you'll actually use on a Tuesday deadline.
|
OpenAI's GPT-5.5 Claims 10+ Hours Saved Weekly — Let's Be Skeptical
OpenAI released GPT-5.5 (they're calling it 'Spud,' which... okay). The headline claim is that early testers saved 10+ hours a week on document review and office tasks. That's a big number. Honestly? I'm skeptical until I see it survive contact with an actual Tuesday production night. But the 'multi-step projects without needing extra user input' piece is interesting. If it actually works — and that's a real if — you could theoretically hand it a 40-page city budget PDF and ask for a summary of the three biggest changes from last year. We'll see.
Worth testing on document-heavy workflows like budget analysis or meeting minutes — but verify those time-saving claims against your own stopwatch before believing the marketing.
Source →
|
The AI Slop Problem in Publishing Is Real — and It's an Opportunity
Paste Magazine ran a piece this week documenting how AI-generated garbage is flooding Amazon. Authors leaving ChatGPT prompts in their unedited manuscripts. AI-generated 'summaries' of books appearing within 24 hours of publication. A survey showing 26% of authors have already lost work to generative AI. It's ugly out there. But here's what I keep thinking: this is actually an opportunity for community newspapers. We know our sources. We were at the meeting. We talked to the family. That's something a slop-generator can't fake. The papers that lean into that authenticity — maybe even publicly commit to it — are building reader trust that matters more every day.
The anti-slop movement is a reader trust advantage waiting to happen — consider an editorial statement about your newsroom's AI disclosure and human-written commitment.
Source →
|
Cloud Computers for Non-Coders: Useful or Overkill?
Manus launched something called 'Cloud Computer' this week — basically a virtual machine that runs 24/7 without you needing to know how to code. You describe what you want in plain English. The use case that caught my eye: a scheduled scraper that monitors local government meeting agendas at 4 AM and delivers a summary before your morning meeting. That's... actually useful? For most weeklies, probably overkill right now. But if you're at a daily or a larger weekly with someone who enjoys tinkering, this is the kind of tool that could automate the stuff nobody wants to do anyway.
Not essential for most small papers today, but worth bookmarking — the ability to run automated monitoring without technical staff is where things are heading.
Source →
|
The Update
• Microsoft's CTO predicts 95% AI-generated code within five years, which tells you where the enterprise players think this is all going.
• Inkitt's CEO admitted AI-generated content has to be disguised as human-written because consumers reject openly AI-made work — a reminder that disclosure and authenticity still matter to readers.
• Meta is facing a class action lawsuit for allegedly using millions of pirated books to train AI models, another reason to watch the copyright space closely.
Small items that didn't warrant full sections but are worth tracking as this landscape keeps shifting.
|
|
|
That 75% number from Google keeps sticking with me. Not because we're going to hit that percentage in newsrooms — we shouldn't, not for the content that matters — but because it shows how fast workflows can change when people actually commit to testing these tools. We are as close in the technology gap today as we are ever going to be. The question isn't whether AI will change how newspapers work. It's whether you'll be one of the publishers who figured it out early... or one who's still waiting for someone else to go first. Trevor Slette runs Quadd.ai — AI tools built for publishers.
|
|
|
|
|
|
From meeting audio to usable copy.
Upload a council meeting, an interview, a coach's post-game scrum. Get a clean transcript, instant summary, and AI Quote Finder that pulls the line you actually need. No more scrubbing through forty minutes of audio at 9pm because the lede quote was somewhere in the middle. Trained on the way reporters actually use audio, not the way podcast editors do.
See how transcription works →
|
|
