Heya and welcome back to Five Things AI!

This week's AI reads reveal a stark divide between executive hype and worker reality: leaders celebrate eight hours saved per week, with 94% trusting AI daily, while 40% of employees would happily ditch it forever, signaling that prompt engineering alone won't cut it. True transformation demands AI vanishing into workflows, unnoticed, especially for SMBs poised to leapfrog giants by tackling tasks they couldn't touch before, unlike enterprises merely streamlining headcounts on familiar plays.

Apple, ever the ecosystem maestro, sidesteps the frenzy, content to distribute frontier models through iPhones without chasing AGI headlines, much like they've funneled others' software for decades.

Deeper in the trenches, MCP and skills spark debates on context pollution, but mastering agentic coding's new vocabulary unlocks real fun, even if integration stumbles for now, with systems sure to sharpen coordination soon.

Claude steals the show regardless, fueling holiday "benders" where coders like Vercel's CTO hack year-long projects in a week for slot-machine thrills. I'm team Antigravity myself, leaning on Claude models laced with Skills and MCP, yet the buzz confirms agentic AI's addictive pull across the board.

Enjoy this edition of Five Things AI!

The AI productivity mirage: Executives see eight hours saved. Workers see almost nothing.

The emotional split runs deeper. Ninety-four percent of executives trust what AI produces. They use it daily. Only 2% avoid it for work. From the top floor, AI is an obvious success.

On the lower floors, people are struggling. Forty percent of all respondents said they’d be fine never touching AI again. That’s not fear of the new. That’s a judgment about whether the thing works. And when these numbers finally reach the boardroom, watch for defensiveness. Watch for doubling down. Organizations that bet this big on transformation don’t reverse course easily.

Interesting findings in these studies. I assume that there are different sets of understanding about AI and what AI can deliver. Telling people to do prompt engineering now as part of their daily routines won’t work, they need AI deeply embedded so they don’t notice it anymore in the workflows. Also, I think that SMB will benefit exponentially from agentic AI because they can suddenly do things they couldn’t do before, whereas large organizations don’t learn new tricks, they can just do what they are already doing more efficiently and with less people involved.

(…continue reading.)

Apple vs. the AI Hype Cycle

Apple sells hardware built around an extremely durable ecosystem. People love Apple products, and even when they don’t love them, they buy them for other reasons. The AI features of a Google Pixel and the foldability of Samsung phones are not enough to make a meaningful dent in Apple’s market share. People will not switch from Apple en masse for AI features alone.

For these reasons, I don’t believe Apple needs to build powerful AI features, at least not in the short term. They just need to keep selling phones.

Frontier AI labs will continue to churn out better models. Meanwhile, Apple will be well positioned to distribute AI, just as they’ve distributed every other software company’s products for decades.

Indeed, Apple has been notably absent from the AI race and Siri still lacks real functionality besides setting a timer (much like Alexa), but I do think that Apple will just blend AI into its products to make them better without confusing the customers too much with talk about AGI.

(…continue reading.)

MCP, Skills, and Agents

I want to believe that people have good intent, but it doesn’t help anyone if you run around exclaiming “MCP is bad” without any nuance around “actually this MCP implementation is bad”. Both skills and MCP - when used incorrectly - will cause context pollution in your harness. It’s important everyone learns the fundamentals of how all of this works, and with the state of technology today that means managing the context window.

Coding with modern agentic coding environments is a lot of fun, but you have to learn a new set of vocabulary first…

(…continue reading.)

Agentic AI and The Mythical Agent-Month

The core premise of "Scalable Agency" seems to rely on a flawed assumption, that software engineering is an embarrassingly parallel task. My intuition, and the paper's own results, suggests the opposite.

I am not so sure about the argument. I understand that integration currently is an issue, but I assume that agentic systems will get better at that as well.

(…continue reading.)

Claude Is Taking the AI World by Storm, and Even Non-Nerds Are Blown Away

Many coders spent their holiday breaks on a “Claude bender,” testing out the capabilities of the latest Anthropic model, Claude Opus 4.5, which they used within a desktop coding tool called Claude Code. Tech companies have been incorporating code-writing AI into their workflows for years, and prior models were often compared with a junior software developer. The buzz around Claude’s latest incarnation is something different.

Malte Ubl is chief technology officer at Vercel, which helps develop and host websites and apps for users of Claude Code and other such tools. He said he used the tool to finish a complex project in a week that would’ve taken him about a year without AI. Ubl spent 10 hours a day on his vacation building new software and said each run gave him an endorphin rush akin to playing a Vegas slot machine.

I am not using Claude Code, I’m in the Antigravity camp, but I’m mostly using the Claude models there. And now Antigravity also uses Skills and MCP and all that good stuff the people at Anthropic thought of. So cool.

(…continue reading.)...