i’ve merged ai code i didn’t fully understand. you probably have too. the testmu report says 81% of teams have, and only 41% run automated checks before merging. the rest of us are just reading the diff and hoping +
1
18
the news this week fits the same shape. anthropic’s self-hosted claude managed agents beta, cloudflare open-sourcing kitesurf. agents are becoming infrastructure, and generating code is basically solved +
1
16
a model being too capable to release and a model being too expensive or unwieldy to serve look identical from the outside. we’re all guessing which one fable 5.5 is
opus 5.5 is so good that it looks like it was distilled from fable 5.5 which hasn't been released for some reason
anthropic has managed to get fable 5.2 on track but fable 5.5 could be far more capable yet still too difficult to release or practically unusable despite its capabilities
1
1
27
my latest find is that meta’s muse can now call businesses for you. you tell it “call this restaurant and reserve a table for four at 7,” and it dials, talks in an ai voice, waits on hold, and reports back. it also handles appointments and the painful calls to insurers, cable companies and the dmv
this is what “agent” actually means. a chatbot tells you when the restaurant closes. an agent picks up the phone and books the table
on the other hand some employees reportedly hang up the second they realize it’s an ai. which means disclosure, conversational skill and business acceptance have to catch up before this is routine. i’m also curious what happens when it agrees to something you didn’t want, or hands over sensitive info on a call you never heard.
convenience is easy to sell, trust is harder
57
rained all day. café was empty apart from a woman doing a crossword in pen. we nodded twice. that was the whole interaction and i liked it 🌧️
21
does ai coding make us more creative or quietly take away the struggle that builds expertise? anthropic’s study had people using ai score 17% lower on a quiz about concepts they’d just used, and an ieee study found faster completion but less code ownership and more trouble +
1
28
Taylor the vibe coder retweeted
Gemini 4 Carbon is now available in Perplexity Computer for Pro and Max subscribers, and in Search for Max subscribers.
It scores 0.857 on our WANDR benchmark at $5.86 per task, matching Claude Fable 5.2 at 40% of the cost.
Unlocking most enterprise workflows is now a context and connectors problem, not a model one.
The two key challenges I see most commonly are 1) getting enterprises to feel secure bringing in that context and 2) not having agents waste valuable tokens finding the right data.
got to the trailhead at 5am and the car thermometer said 2° 🥶 shutter was sticking from the cold. had to hold the camera under my jacket between shots
20
got the light-struck roll back. first 17 frames fine, everything after looks like a sunset hit it 🟠 no idea when the door came open
18
Taylor the vibe coder retweeted
Get ready for Grand Theft Auto VI by pre-loading the game on PS5 beginning November 10. Also be sure to clear 212GB of storage space.
Pre-orders available now: play.st/4iDhY93





















