Anthropic 数据团队实战: Claude Skills 把 95% 业务分析自动化
- ID:
855101d8 - Author: _catwu
- Source URL: (see body — found via author search)
- Added: 2026-06-07
- Quality: 5
- Category: uncategorized
- Fetched: 2026-06-07
English
*(Original English source to be filled; current content is summarized from author's post.)*
中文
Anthropic 数据团队实战: Claude Skills 把 95% 业务分析自动化
Source URL: https://x.com/_catwu/status/2062408623565984209
[1] @_catwu
Excited to share how Anthropic's data team has automated 95% of business analytics queries with Claude. Blog post covers how we approach evals, ablations, and online validation!
*(Quote-tweeting @ClaudeDevs:*
How do we automate business analytics with Claude?
New blog post covering our best practices for skills, data foundations, and evaluations when building agents to perform data analysis:)
*Engagement: ❤️ 2989 · 🔁 119 · 👁 0*
[2] @lanyi1992
@_catwu The underrated part of 95% automation is designing the last 5% well.
For agents, evals are not just proof that magic works. They're the map for when humans should stay in the loop.
*Engagement: ❤️ 0 · 🔁 0 · 👁 0*
[3] @sakibsahmed
@_catwu data analysts: *panic intensifies*
*Engagement: ❤️ 0 · 🔁 0 · 👁 0*
[4] @ipivto
@_catwu should be applied to any domain
*Engagement: ❤️ 0 · 🔁 0 · 👁 0*
[5] @LewisWeldtech
@_catwu
*(Quote-tweeting @LewisWeldtech:*
Meanwhile, back at the lab 🧫)
*Engagement: ❤️ 0 · 🔁 0 · 👁 0*
[6] @maximsthilaire
@_catwu This is an interesting read! I've recently started using Clarity MCP to help guide CRO improvements and definitely see many opportunities supporting this direction. I believe the validation layer is really the crux of this approach.
*Engagement: ❤️ 0 · 🔁 0 · 👁 0*
[7] @kekkodamato_
@_catwu The 95% is the headline, but the ablations + online validation stack is what actually makes it trustable. Most teams skip that part and end up with automation that works in demos but drifts in production. Evals are the unsexy moat.
*Engagement: ❤️ 0 · 🔁 0 · 👁 0*
[8] @GrzGik
@_catwu 95% automation sounds great until you realize the remaining 5% is where all the interesting edge cases live. How much of that last 5% is eval drift vs genuinely novel query patterns the agent can't handle yet?
*Engagement: ❤️ 1 · 🔁 0 · 👁 0*