营销工作者的 AI 模型分工：Astra 负责电脑操作与 3D
一份模型选型清单，把 Astra 定位在浏览器自动化、3D 建模和跨 agent 项目编排，Codex 与 Claude 各管一段。
my AI model stack for marketing work 1. Codex / GPT-5 series > copywriting when you give it a solid writing voice harness > coordinating campaigns across several channels and agents > turning research into detailed briefs, campaign plans and channel-specific tasks > logical work that has to follow a fixed format or pass a checklist 2. GPT-6 Astra > computer use for browser workflows where no API or MCP is available > automating repetitive work inside tools such as LinkedIn > 3D modelling and turning references into working scenes > orchestrating larger projects across agents and tools > reviewing work between stages and catching quality drops before the next agent continues > keeping the original brief, constraints and quality bar consistent across a long project 3. Claude Fable series > front-end development and design > reviewing pull requests and planning how the database, agents and front end connect > architecting marketing databases and data warehouses > mapping possible solutions when I have a problem but have not decided what to build Fable is the model I open when I still do not know how the system should work 4. Claude Opus & Sonnet series I find Opus almost worthless on its own, but it becomes a proper workhorse as a subagent managed by Fable or GPT 6 Astra > implementing a plan that has already been decided > working through a backlog > making changes across a large codebase > taking scoped tasks from the main agent and returning the files for review 5. Grok series > research that needs current X context > finding the language, opinions and conversations around a topic > adding more personality and fun to copy > producing another front-end concept when I want a different visual direction I like pairing Grok with Codex for content. Codex writes the outline and first draft, Grok adds context and makes the writing feel more personal 6. Kimi K3 > long research jobs that need many sources combined into one clear report > front-end development > executing scoped tasks as a subagent managed by Fable or GPT 7. GPT Image 2 > polished mockups > clean text inside images > diagrams, ads, social assets and presentation visuals > changing one part of an image while keeping the composition > working inside Paper to quickly generate landing-page mockups and other visual assets this is my go-to image model right now 8. Seedance 2.5 > 30-second campaign videos with connected shots > keeping the same product, character and art direction across a sequence > building from image, video and audio references > editing a specific moment by timestamp instead of regenerating the full video > product demos, launch films and paid social creative
这条是引用着另一条写的：x.com/1455999313453195265/status/209470418334592…