⦿ Field Notes 5 // ChatGPT Atlas, Claude café, and tech optimism

For every AWS outage, a breakthrough in home robotics

Author

Matt C. A. Smith
November 03, 2025

Welcome back to Field Notes.

I'm feeling bullish on technology in a way I haven't since the mid-2010s. I'm still not fully convinced on the AI non(?)-bubble, but it feels like the sector has regained some of its soul recently. At least in my circles, there's been pushback against bland tech proclaiming to change to world, and a refocus on the humans who use it.

This was kickstarted - or at least catalysed - by the Claude pop-up café Anthropic ran in New York. I narrowly missed it, but even from a distance it was encouraging to see colour, rough edges, and above all, optimism. Anthropic gets the concerns around AI, and has positioned Claude as an assistant to support human thinking, rather than a replacement for it. Of course, it helps that it's also a very good LLM.

The reaction to the Friend AI pendant was fascinating to watch. Its advertising campaign was graffitied and reviews were... less than complimentary. Even ignoring the ethical and philosophical implications of a device claiming to offer friendship akin to the real thing, it's amazing nobody thought it was a bad idea to push alerts every two minutes to say that Friend can't understand what it's hearing.

And rounding off the month, I was stunned at 1X’s demo video for the NEO robot. I wouldn’t buy one at this point - having operators remote in to my home to perform tasks is a dealbreaker for me - but a $500-per-month robot assistant who can take care of my home chores sounds like a fantastic value proposition to me, and much closer to the future we were promised. I can’t wait to see how it develops.

This month on the blog

ChatGPT Atlas positions AI as the internet's gatekeeper

We’re shoehorning AI into everything else, so why not your web browser?

mattcasmith.net/2025/11/02/chatgpt-atlas-ai-browser

Claude's café pop-up convinced me to switch LLMs

There’s a new cool kid on the AI block, and its name is Anthropic

mattcasmith.net/2025/10/05/anthropic-claude-ai-marketing-design

EtherHiding: When good blockchains turn bad

Threat actors are uploading “bulletproof” malware to the blockchain

mattcasmith.net/2025/10/26/etherhiding-blockchain-malware

The untapped potential of AI for security operations

AI needs to take another step forward before it provides day-to-day value in the SOC

mattcasmith.net/2025/10/19/ai-security-operations-untapped-potential

Things that left an impression

Field notes from the world of technology and design:

  • The recent AWS outage exposed yet more weaknesses in the ever-growing Internet of Things. Cloud-powered smart bed Eight Sleep kept users awake when the device failed to call home, with some devices stuck in an upright position or at uncomfortably warm temperatures.

  • Speaking of AWS, software engineer Jonathon Belotti wrote a fantastic blog post covering the reasons for the outage, explaining how a DynamoDB issue cascaded into a major incident affecting an array of services (yes, it was DNS - but the full story is far more interesting than that).

  • Founders podcast creator David Senra recently launched a second series in his own name, and Spotify's Daniel Ek stopped by for the first episode. One of many interesting tidbits was that Sony's founders hired a "paid critic" to identify deficiencies in their products. When you're too close to something you're building, you can miss the forest for the trees.

  • OpenAI alumnus Andrej Karpathy spoke to Dwakesh Patel about AI agents, labelling them “cognitively lacking” and predicting that it might take a decade to solve current issues and reach the nebulous concept of AGI.

  • And Ed Zitron recently dropped some stunning AI industry figures in his newsletter. He reported Anthropic spent $2.66 billion on AWS compute between Q1 and Q3 2025, against revenues of $2.55 billion - that's $1.04 spent for every $1.00 earned, without even considering other costs!

That’s all for now - I’ll see you next month!