All About AI Automation

AI Inside for Wednesday, January 21, 2026

This episode of the AI Inside newsletter is sponsored by Your360 AI. Get 10% off through January 2026 when you use the code: INSIDE.

Lot’s of interesting stuff this week on the AI Inside podcast, including Anthropic’s new Claude Cowork agent, OpenAI’s ad-supported ChatGPT shift, YouTube’s war on “AI slop,” and a deep dive into real-world AI automation with Alfred Nutile. And that is just scratching the surface of this episode.

But first, big thanks to this week’s Patron of the Week: Roland Aicele, Gabrielle Haley, and Tudo!

Claude Cowork’s new agentic desktop powers

Claude Cowork launched as a new agentic feature in the Claude macOS app that lets Claude read, edit, and create files in a folder you choose. So imagine organizing your desktop, or assembling random notes into structured docs or spreadsheets, or running multi-step tasks with progress updates along the way. Users assign folder permissions and also have control via confirmation before major actions. Anthropic says they’ve built in protections against prompt injections, but also warns that more work needs to be done there. You’ve been warned. First released as research preview for Claude Max subscribers ($100-200 per month), then expanded to include Pro plan users ($20/mo).

OpenAI brings ads to ChatGPT

OpenAI introduced an ad-supported model for ChatGPT’s free and Go tiers. Go is the tier that costs $8 per month which, by the way, also now includes unlimited GPT-5.2 Instant access. Pro, Business, and Enterprise plans remain ad-free. Those ads will appear with labels and separated from answers, testing first for logged in US adult users. Users can turn off personalization. The Information reported that OpenAI is actively offering those ads to dozens of advertisers, charging based on ad view, not ad click in the early stages. OpenAI may be one of the first to actually serve ads to LLM content, but they are all flirting with this idea whether they admit they are or not.

YouTube’s war on “AI slop”

YouTube CEO Neal Mohan published his annual letter to the YouTube community, and he had some choice things to share about “AI slop.” In it, he likens this moment of AI content to previous moments of content that once puzzled the mainstream but now is accepted, like ASMR and Let’s Play streams. He wrote that to reduce the spread of low quality AI content, YouTube is actively building on its systems for spam, clickbait, and repetitive content, while still supporting a broad range of free expression. He is also pushing to crack down on deepfakes and protect likenesses, while YouTube rolls out more AI-powered creative tools like Dream Screen, auto-dubbing, and Shorts tools that let creators use AI versions of their likeness.

“Vibe coding” and the rise of personal micro apps

TechCrunch had a story last week that takes a closer look at “vibe coding.” What caught my attention was that this article doesn’t look at vibe coding from the usual perspectives of developers using AI to help them create portions of code or non-developers vibe coding to “become a coder.” This focused on everyday people choosing to vibe-code apps tailored to their specific needs right now instead of hunting for something in the app store. These apps are not necessarily meant to become successful consumer products, but solutions to personal needs. Call them micro apps, personal apps, fleeting apps. Hyper context specific, niche by design, and often they “disappear when the need is no longer present.”

Alfred also brought up Opal by Google, an AI service that is really all about easy access development of these low-lift AI apps and its pretty darn cool!

Alfred Nutile’s real-world AI automation playbook

Speaking of, because Alfred Nutile on the show, I thought it would be a cool opportunity to talk with someone who builds AI automation systems for clients for a living about what that’s like. Alfred is an AI & automation developer and architect at DailyAI and YouTube creator on AI automation. We break down how so much of AI is about using tools to make lives, jobs, and workflows easier, and why consistency is the real challenge. Alfred shares his day-to-day helping clients automate tasks like turning handwritten notes into structured data via photo scans or gathering website data into CSVs without heavy coding.​

He explains choosing tools through trial and error, like testing Claude vs. Manus for depth, treating AI as “salt and pepper” on top of automation basics like Lego-block workflows. We discuss if dev background helps (it speeds things up but isn’t essential), low-hanging fruit everyone can grab (automation tools for repetitive tasks), mapping client needs to avoid overcomplicating, and how he turns real builds into YouTube content. Communication is the hardest part, but tools like Claude Cowork are making it practical for non-devs. Lot’s of ground covered here!

AI’s water use vs burgers

I came across an analysis piece called “From Tokens to Burgers: A Water Footprint Face-off.” In it, the authors argue that the water panic around AI datacenters ignores scale and context. It uses Elon Musks’s 400 megawatt Colossus 2 site as the AI example, saying it uses about 346 million gallons of blue water per year. It compares that to the fast food chain In n Out, counting only their Double Double burgers, and says a single restaurant comes in around 147 million gallons. In essence the article argues that one of the largest AI data centers equals roughly 2.5 burger joints. I am DEFINITELY over simplifying the article, but that comparison caught my attention.

OpenAI’s age-prediction and “adult mode” prep

OpenAI is officially introducing its age-prediction system in ChatGPT worldwide. It is intended to spot users under the age of 18 and add stronger protections on those accounts. The system will estimate the age of a users account based on chat history and other indicators and apply restrictions that limit exposure to sensitive content when triggered. Adults who are misclassified will be able to restore full access by submitting a selfie through verification provider Persona. Yes, this is happening ahead of a possible roll out of an “adult” mode sometime later this quarter, at least according to OpenAI last December.

Adobe’s new Premiere AI tools

Adobe is bringing more AI features to Premiere that have me excited. New masking, tracking, and motion design features become one-click operations instead of old-school keyframing. The new Object Masking tool lets editors hover and click on a person or object to auto-generate a colored mask overlay that tracks throughout the clip, generated on-device with no data sent to Adobe. Those masks can be adjusted and tweaked afterward. There is also an easier import path for objects generated using Adobe’s Firefly system into the apps, which should speed up creative workflows.

Lego’s AI learning kits for kids

Lego’s new Computer Science and AI Learning Solution kit is a thing. It is based on its recently announced Smart Play system from CES a few weeks ago. It uses locally run tools to teach K-8 students core computer science concepts, including machine learning, computer vision, and generative models. None of it is chatbot driven. Pre-order for the K-2 sets starts at $339.95, with older kits at $429.95 and $529.95, shipping in April.

Wikipedia’s big AI licensing deals

Wikipedia celebrated its 25th birthday by striking a deal with Amazon, Meta, Microsoft, Mistral, and Perplexity. It is licensing Wikipedia’s data for training models. Google, Ecosia, Nomic, Pleias, ProRata, and Reef Media were already part of the program. This provides API access to those licensees and far more capabilities than the free API. Wikipedia reported last April that bandwidth had risen 50% since January 2024 as more companies scraped Wikipedia for training data.

HUGE thank you to Executive Producers on the Patreon: DrDew, Jeffrey Marraccini, Radio Asheville 103.7, Dante St James, Bono De Rick, Jason Neiffer, Jason Brady, Anthony Downs, Mark Starcher, and Karsten Samaschke!!

Thank you! See you next time on another episode of AI Inside.