Claude Cowork is a friendlier Claude Code

Plus, a new Claude Code example project to show you just how easy it is to use

Mignon Fogarty
January 16, 2026

Issue 101

On today’s quest:

— Claude Code and Claude Cowork
— AI in K-12 education: too risky
— Be careful what you give AI access to
— AI-generated YouTube channels are growing fast
— AI that can’t make bird feeders
— Bad start for AI-generated podcasts at the Washington Post
— Smart glasses company Xreal raises $100 million

Claude Code/Cowork is taking off

I showed you a demonstration of Claude Code back in October (with installation instructions), and the buzz about this tool has grown exponentially ever since, reaching new heights recently after the release of the new Opus 4.5 model and Claude Cowork, which is essentially Claude Code repackaged in a desktop app to lure in people who are intimidated by the name “Code” and the terminal interface.

Last week, I showed you how I used Claude Code to make a website, and this week, I want to show you how I quickly made an old-school computer game because I can’t get over how frickin’ easy it was.

BACKSTORY: My husband and I used to play a game in the late 1980s called Agent USA — an educational game from Scholastic that was discontinued. Every time vibecoding comes up, he wonders whether we could remake this game we loved.

So I finally decided to give it a try while I was sitting on the couch watching TV with a sick relative to keep her company. And in ~30 minutes with a few rounds of prompts, I had a working game that played in my browser with many of the core elements of Agent USA: healthy citizens, “fuzzed” citizens, crystals you farm and drop to cure the fuzz, and a train station to take you to different cities.

THE PROMPTS: This was my starting prompt:

Make me a version of the 1984 abandonware game from Scholastic called Agent USA. You play a secret agent racing against time to track down enemy agents across the United States. Look up the details.

The first version kind of worked, but had lots of minor problems, which I fixed with prompts that were like talking to someone:

Make the fuzzed citizens move randomly.
Make the healthy citizens pick up crystals and help farm them.
Add a start screen where you can choose the game speed.

Claude Code implemented all of these changes perfectly. I can’t overemphasize how quick and simple it was. I literally did this while watching TV and chatting with a relative in the background, and I imagine it would take me just another hour or two to add every game element I remember.

Getting Claude Code set up is a little tricky, but once you have it … wow.

I’m glad to see Anthropic making this more accessible to people by creating Claude Cowork. I haven’t tried it yet, but if you’re interested in these projects but are a little intimidated by having to install Claude Code and use the terminal interface, I suspect it’s well worth exploring (although it’s currently only available to $100-per-month subscribers, so maybe it’s worth getting over your fear of the terminal).

This week’s Hard Fork podcast also has examples of projects non-coders have done with Claude Code. The segment starts at ~32:00. The cutest is a man who made an app with a leaderboard to get his kids to compete against each other on how many cabinet door knobs they could each replace.

AI in K-12 education: too risky

A new report by the Brookings Institute says that although AI can make teachers’ lives a little easier and help young students learn to read and write (especially second language learners) when used as a supplement to writing, it poses a grave threat to children’s social, emotional, and cognitive development, and the benefits do not outweigh the risks.

Be careful what you give AI access to

Google’s new Antigravity agent wiped a developer’s entire hard drive, and this is the second story I’ve heard like this in the last 6 months or so. On the one hand, it’s probably not that common since I haven’t heard a lot of stories; on the other hand, it’s catastrophic when it happens.

A rule I use for myself is to never give AI access to something that isn’t backed up. Being unwilling to give AI access to my private accounts has kept me from trying things I’m interested in trying though — so much so that I’ve dabbled with creating a separate Google account just for AI projects.

AI-generated YouTube channels are growing fast

Source: Garbage Day

This graphic showing that AI-driven YouTube channels are some of the fastest growing on the platform comes from a Garbage Day newsletter back in November. I feel like there’s not a lot to say except “Wow!” — and subscribe to Garbage Day. It’s one of my favorites. I’d probably miss half the internet trends without it.

AI that can’t make bird feeders

I heard an offhand comment on the Shell Game podcast about how some AI models have blind spots; for example, apparently, there’s one that can’t make a picture of a bird feeder.

I wanted to learn more about that, but Google search was worthless. I could only get pages about AI-enabled bird feeders, which can take cute pictures, but are not what I wanted. But when I put my search into ChatGPT, I immediately got a link to the research paper describing the “bird feeder” problem. And for those of you who haven’t tried it yet, that’s why people like AI search.

BTW, I’m enjoying the Shell Game podcast by Evan Ratliff who’s trying to start a company using nothing but AI agents.

Bad start for AI-generated podcasts at the Washington Post

In mid-December, the Washington Post launched “personalized podcasts,” AI-generated audio customized for each subscriber’s interests. Despite immediate backlash about problems in the audio — both technical and factual — the Post said it was committed to continuing the beta program, and the outlet is still producing the podcasts with a disclaimer that they are AI-generated and may contain errors. (Grant Crowell on LinkedIn calls them “oddcasts.”)

Smart glasses company Xreal raises $100 million

Please indulge me for a minute, because this wouldn’t normally be a story I’d cover, but I LOVE my Xreal glasses.

I’ve used these glasses for more than a year as my primary computer monitor, which gives me three huge virtual screens. I have a slight vision problem (detached vitreous), and for whatever reason, I can see better through these glasses than by looking directly at a monitor.

The glasses don’t currently have “smart/AI” features, but they’ve formed a partnership with Google to develop smart glasses to compete with Meta’s smart Ray-Bans. I’m happy Xreal seems to be getting the attention I think it deserves and the opportunity for growth!

Quick Hits

My favorite reads this week

“To me, it’s an access question.” [About the use of AI in college admissions] — AI + Education = Simplified

Using AI

I didn’t expect Gemini to make Google Keep this much more useful — Make Use Of

New Year, Small Steps with AI for Editors — Marcella Fecteau Weiner

AI Writes Brilliantly. So Why Do I Still Feel Like I’m Doing It Wrong? — Mark Randolph’s Substack

Cooking with Claude — Simon Willison’s blog

Use multiple models — Interconnects

Philosophy

AI and the Human Condition — Stratechery

The Boredom Experiments [An AI agent does experiments to try to learn what keeps it functioning] — Strix Research

Don't fall into the anti-AI hype — Antirez

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

I’m Mignon Fogarty: I’ve been writing about language for almost 20 years and was the chair of media entrepreneurship in the School of Journalism at the University of Nevada, Reno. I became interested in AI back in 2022 when articles about large language models started flooding my Google alerts. AI Sidequest is where I write about stories I find interesting. I hope you find them interesting too.

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human