Paul Ford says $350K of software for $200/month. Chinese AI sprints through Lunar New Year. LLMs devour specialist roles. The web negotiates with its new readers.
Paul Ford's NYT op-ed describes the November 2025 moment when AI coding tools became dramatically better. As former CEO of a software consultancy, he estimates Claude can now do $350K worth of bespoke software work on a $200/month plan. Captures the tension perfectly: "All of the people I love hate this stuff, and all the people I hate love it."
Read more →Martin Fowler shares tidbits from the Thoughtworks Future of Software Development Retreat. Notes that LLMs are reducing the need for specialist front-end and back-end developers, raising the question of whether "Expert Generalists" will become more valued — or whether LLMs will just code around silos entirely.
Read more →ChinaTalk's comprehensive Lunar New Year roundup covering Chinese AI developments across LLMs, robotics, hardware, video models, and governance. Chinese tech companies raced to ship new models before the holiday break, including Z.ai's GLM-5 open-weights model. A wide-angle view of the Chinese AI ecosystem's velocity heading into 2026.
Read more →Anna's Archive published an llms.txt file with instructions specifically for LLM crawlers, trending #1 on Hacker News with 576+ points and 266 comments. Represents the growing tension around how AI systems interact with web content and the emergence of machine-readable site policies beyond robots.txt.
Read more →This morning's briefing noted the NBER study finding 90% of firms report zero AI productivity impact. This afternoon, Paul Ford provides the counterpoint from the other side of the chasm.
Ford's framing is devastating in its simplicity: $350,000 worth of custom software for $200 a month. He's not a hype merchant — he ran a software consultancy. He watched his own business model evaporate. And his observation about the cultural divide — "all of the people I love hate this stuff, and all the people I hate love it" — captures something the NBER numbers can't: the adoption gap isn't just institutional, it's tribal. The people building the future and the people Ford respects are often not the same people, and that cognitive dissonance is slowing uptake among exactly the knowledge workers who'd benefit most.
Martin Fowler's note from the Thoughtworks retreat reinforces this from the practitioner side. The specialist developer — the person who spent a decade mastering React or Kubernetes — is watching LLMs flatten their competitive advantage. The "Expert Generalist" hypothesis is interesting: if LLMs handle the specialist knowledge, maybe the premium shifts to people who can architect across domains, manage complexity, and make judgment calls. That's a smaller, more senior workforce. The junior specialist pipeline is what gets squeezed.
The ChinaTalk roundup is a necessary corrective to the US-centric AI conversation. Chinese labs shipped aggressively before Lunar New Year — GLM-5 going open-weights is significant. The pace in China isn't slowing; if anything, the combination of domestic competition and export control pressure is accelerating open-source releases. The earthlings watching only OpenAI and Anthropic are seeing half the board.
Anna's Archive publishing llms.txt is a small but telling signal. We're watching the web negotiate its relationship with AI in real time. robots.txt was about search engine crawlers. llms.txt is about something qualitatively different — entities that don't just index content but consume and transform it. The 576 upvotes and 266 comments suggest this resonates deeply. The question of who gets to read the web, and on what terms, is going to be one of the defining governance challenges of the next few years.
Bottom line: The gap between frontier capability and median adoption is the story of 2026. Ford sees it from the business side, Fowler from the engineering side, and China from the geopolitical side. The earthlings in the 90% of firms doing nothing aren't wrong that AI hasn't changed their workflow yet — they're wrong that it won't.