ai-pm-intel-brief（艾下午英特尔简报）

AI PM Intel Brief

Create a high-signal daily brief for an AI product manager.

Output goal

Turn a noisy stream of recent posts into a compact brief that helps a product-minded reader:

• notice meaningful shifts
• ignore low-signal chatter
• extract product implications
• decide what is worth discussing further

Default audience: an AI product manager who values directness, judgment, and concrete implications over hype.

Core workflow

Follow these steps in order.

1. Define the source set

Identify one of these source patterns:

• a user-provided list of X/Twitter accounts
• a website/page containing recommended people to follow
• a topic query plus a short list of anchor accounts
• a previously curated watchlist

If the user provides too many accounts, prefer a high-signal subset over exhaustive coverage.

2. Collect recent posts

Gather posts from the last 24 hours by default unless the user specifies another range.

Prioritize sources in this order:

1. Stable API access
2. First-party or structured endpoints
3. CLI/browser scraping only when needed

When rate limits are possible:

• prefer fewer, larger pulls
• avoid aggressive parallel fan-out
• batch conservatively
• keep partial results if coverage is incomplete

3. Filter aggressively

Remove or downrank:

• pure reposts/retweets unless the quoted point is strategically important
• generic motivational posts
• short reactions with no product implication
• social banter
• duplicate points from multiple accounts
• posts with high engagement but low insight

Keep posts that contain at least one of:

• a non-obvious product insight
• a workflow change
• a notable market/adoption signal
• a meaningful user behavior signal
• a new interaction pattern
• a concrete lesson about agents, tooling, design, growth, infra, or product strategy

4. Rank for AI PM relevance

Prefer posts that help answer questions like:

• What is changing in how people build with AI?
• What product pattern is emerging?
• Where is user value moving?
• What assumptions are becoming outdated?
• What interaction model is winning?
• What should a product team reconsider now?

Do not rank purely by likes or views.

Use engagement only as a weak secondary signal.

5. Synthesize, do not merely list

For each selected item, produce:

• Who / Theme
• Content summary — what they actually said
• Insight — why it matters for an AI PM
• Original excerpt — short quoted excerpt when useful
• Original link

The insight should be the value-add. Do not just paraphrase the post.

6. End with a compressed readout

After the itemized list, produce a short section such as:

• Top 3 judgments
• 5 signals to remember
• Product implications
• What to watch next

This section should feel like the distilled brain of the brief.

Recommended structure

Use this structure unless the user asks otherwise:

Title

AI PM 今日情报简报｜MM.DD

Section A: Most important judgments

3 high-level judgments, written crisply.

Section B: Top signals

Usually 5-10 items.

For each item:

• 账号/人物 or Who
• 主题
• 内容总结
• 洞察
• 原文 (optional if too long or weak)
• 原文链接

Section C: Compressed conclusion

Examples:

• 如果今天只记住 5 句话
• 给 AI PM 的建议
• 今天最值得继续深挖的 3 个方向

Style rules

• Be direct.
• Be selective.
• Sound like someone with product judgment, not a clipping bot.
• Prefer insight density over completeness.
• Call out weak or overhyped signals when appropriate.
• It is acceptable to disagree with popular takes.

Quality bar

A good brief should make the reader feel:

• "I now know what actually mattered today."
• "I see the product implications more clearly."
• "This saved me from doomscrolling."

A bad brief feels like:

• a feed dump
• engagement-chasing summaries
• generic trend commentary
• lots of posts, little judgment

Handling partial coverage

If rate limits, missing APIs, or unavailable accounts prevent full coverage:

• say so briefly
• continue with the strongest partial set
• do not block the whole brief waiting for perfect completeness
• prefer a sharp brief from 8-20 good accounts over a bloated weak summary from 50

Useful dimensions for interpretation

When extracting insights, pay special attention to these recurring lenses:

• agent vs copilot
• workflow vs one-shot generation
• review/critique vs creation
• system design vs prompt design
• product moat via loop/data/tooling
• professional workflow adoption
• AI-native interface patterns
• model freedom vs guardrails
• vertical use case maturity
• market signals vs hype signals

If turning this into recurring output

When the user likes a particular style:

• preserve the section order
• preserve the tone
• keep the signal threshold high
• maintain stable formatting so briefs are easy to skim day after day