一周四款个人助理更新：Dots、GrokBot、Sonnet 5.5、Gemini 4

s1rozha1 @s1rozha_原推发布于 2026.10.07

在 X 上打开原推

AI AGENTS JUST HAD THEIR “iPHONE MOMENT” one week, four major shifts: > Anthropic dropped Sonnet 5.5 > OpenAI launched Dots > Google teased Gemini 4 Argon > GrokBot added real agent-team upgrades the biggest one is OpenAI Dots. it’s basically a personal agent inside ChatGPT that can talk by voice, use your apps, spin up Codex sessions, write docs, send Slack messages, send emails, and work across GPT Space. real workflow from the video: > 35-minute voice call with the agent > agent drafts email + Slack update > sends both after approval > spins up a local Codex session > starts redesigning a landing page > creates 3 landing page options > pushes work into Sites for preview that is not “chatbot productivity.” that is agent orchestration. OpenAI also added GPT Space: > pages > sites > images > Google Drive > subpages > AI-editable docs > agents that can write into your workspace basically Notion + Google Drive + agent memory, inside ChatGPT. then there’s GPT 6.1 Soul: > 5x cheaper than Astra > strong in real usage > fewer limit issues > usable inside the new ChatGPT app workflow meanwhile Anthropic dropped Sonnet 5.5 just 6 days after Opus 5.5. the weird part: Sonnet 5.5 is reportedly better at agentic coding than Opus 5.5 on their chart. and it’s half the price: > Opus: $4 in / $20 out per 1M tokens > Sonnet: $2 in / $10 out per 1M tokens but the creator’s take was clear: Opus 5.5 is still the monster model. especially inside Claude Projects, where one main agent can delegate into separate threads for app builds, decks, docs, and cloud/local sessions. then Google enters with Gemini 4 Argon. reported claims: > new frontier model > cyber defenders get access first > $2 in / $10 out > wider release “as soon as possible” but the skepticism is fair: Google has had strong model cards before, but weak real-world agent workflows. the missing piece is not just the model. it’s the app. ChatGPT has the desktop app. Claude has Claude Projects. GrokBot has the agent team interface. Google still needs one obvious “use this” AI super app. and GrokBot quietly shipped useful agent updates too: > proactive suggestions > inline email drafts > inline Slack drafts > team bots in Slack > route traffic through your computer that last one matters because cloud VMs get blocked. routing through your own computer makes the agent look more like you when browsing, buying, or filling forms. AI agents are moving from “answer my prompt” to: > talk with me > remember my workspace > write into my docs > message my team > launch coding sessions > route work to specialist agents > use my computer when the web blocks bots bookmark this before agent apps get impossible to keep up with

这条是引用着另一条写的：x.com/s1rozha_/status/2066919886396244101