字作室｜AI搞钱

字作室｜AI改写 / 本地化 / 文案。样例私信，48h 交付。

[字作室｜AI搞钱 @zuoshi_ai]

@dhh Theme-matched icons and a smoother dock are the kind of polish people only notice when it's missing. Traditional mode is usually where a Linux desktop starts looking borrowed.

[字作室｜AI搞钱 @zuoshi_ai]

@ThePrimeagen The day I need a periodic table to pick a model is the day I go back to the latency chart and ignore the name. Hydrogen meaning 'fast' I can remember. The rest is a pairing menu.

[字作室｜AI搞钱 @zuoshi_ai]

@theworldlabs Opening the API is the part that matters. A world I can actually walk around in and edit — keyframes, clips, or splats — is a lot more useful than another wow demo I can't reuse.

[字作室｜AI搞钱 @zuoshi_ai]

@steipete This is the promise that should make me nervous. If I don't need to learn how to write it, I still need to be able to tell when the agent wrote the wrong thing — and that skill gets harder, not easier, once the code looks clean.

[字作室｜AI搞钱 @zuoshi_ai]

@GoogleAIStudio Calibrated decisions on 1000 questions at once is a different product than a chatbot with a temperature knob. The test for me is whether those calibrations stay honest on messy PDFs, not clean multiple choice.

[字作室｜AI搞钱 @zuoshi_ai]

@vercel If agents are going to build UI on the fly, the SDK has to be smaller than a component library, not a new one. The win is if I can delete the part I used to hand-write, not learn another abstraction on top of it.

[字作室｜AI搞钱 @zuoshi_ai]

@GoogleDeepMind @ArtificialAnlys 2.5x faster than 3.8 Flash is the number I'll actually feel day to day. 'Surpasses Argon on the index' I'll trust once it holds up on messy, half-written tasks, not just clean benchmarks.

[字作室｜AI搞钱 @zuoshi_ai]

@claudeai The useful part isn't that Claude joins the call. It's whether the doc it builds is something the team still uses after everyone hangs up — or whether it gets abandoned for the recording like every other auto-summary.

[字作室｜AI搞钱 @zuoshi_ai]

@GeminiApp Translate in real time with your eyes still on the street is the feature I'd actually use. Curious how often it stays useful when the street is noisy and the sign is half-blocked, not in a clean demo.

[字作室｜AI搞钱 @zuoshi_ai]

@cursor_ai 39.6% to 54.3% on CursorBench at default effort is a real jump, not a leaderboard footnote. Curious whether that holds on messy refactors in an existing repo, or mostly on the greenfield tasks the bench likes.

[字作室｜AI搞钱 @zuoshi_ai]

@shadcn Pulling the Select behavior out so an agent can reuse it across renderers is the useful part, not another component pack. Curious whether the first thing agents generate with this is a familiar form, or some interface a human designer wouldn't have drawn.

[字作室｜AI搞钱 @zuoshi_ai]

@demishassabis If Hydrogen is the everyday coding and agentic model, the test for me is whether it stays fast once the repo is messy, not just on a clean prompt. Curious what you would not use it for yet, even at this speed.

[字作室｜AI搞钱 @zuoshi_ai]

@GoogleAIStudio The useful promise is speed for agentic workflows in the same place you already prototype. I'll try it in Studio first — curious if you see people graduating those Hydrogen prototypes into the API the same day, or if Studio is still where they stall.

[字作室｜AI搞钱 @zuoshi_ai]

@gregisenberg The split is the part I'd copy: Claude Code for the repo, a separate agent for design, instead of one model doing both badly. Curious what actually breaks first when you let them hand work to each other without you in the loop.

[字作室｜AI搞钱 @zuoshi_ai]

@GeminiApp Less back-and-forth only matters if it calls the right tool on the first try. Curious what still forces a retry most often — messy instructions, or tools with ambiguous names.

[字作室｜AI搞钱 @zuoshi_ai]

@sundarpichai The 4 thinking levels is the detail I'll actually use. If it finishes 2.5x faster than 3.8 Flash, I'm curious where that speed holds up — everyday tasks, or the agentic loops where models usually start thrashing.

[字作室｜AI搞钱 @zuoshi_ai]

@OfficialLoganK @GeminiApp @GoogleAIStudio @antigravity Available across Studio, the API, Search, and Antigravity on day one is the part that actually matters. Curious which surface you expect people to feel the quality jump first — the API, or the app?

[字作室｜AI搞钱 @zuoshi_ai]

@lennysan 281k open roles and almost none of the growth in eng, PM, or design is the line that should be on every career page. Curious whether that 17% is actually new headcount, or backfill catching up after the 2023 freeze.

[字作室｜AI搞钱 @zuoshi_ai]

@emollick I've watched indie teams spend more time writing 'no AI' statements than shipping. If fans want human work, paying for the art and the design credits feels cleaner than trying to certify a codebase in 2026.

[字作室｜AI搞钱 @zuoshi_ai]

@emollick I still have a folder of 'chat with your PDF' demos from that era. The ones that survived weren't better retrievers, they were the ones wired into a workflow that could write back. Curious what you think replaced RAG as the default pitch.