Replace your weekly report with AI

The weekly report is one of those tasks where the cost of writing it has crept above the cost of reading it. Most are skimmed in 30 seconds and written in 90 minutes. AI flips that ratio.

Friday afternoon doesn't have to be the day you stare at a blank doc trying to remember what happened on Tuesday. The pipeline below takes about ten minutes once it's set up, and the output is closer to your real voice than what you're writing by hand right now.

The pipeline, in one sentence

Feed the model the raw artifacts of your week. Ask for the report shape your audience wants. Edit the one paragraph that sounds off. Send.

That's it. The interesting part is the artifacts.

The raw material

Your week already produced everything the report needs. You just haven't gathered it in one place. The usual mix:

• This week's calendar events, copied straight out of Google Calendar or Cal
• Slack DMs and the channels you actually post in (export the last seven days)
• Git commits, Linear tickets, or whatever your project tracker calls "done this week"
• Meeting transcripts from Granola, Fathom, or Read.AI
• Your own scratch notes, even the messy ones

You don't need all of these. You need the two or three that capture where your week actually went. For an engineer that's commits plus Linear. For a PM that's meetings plus Slack. For a founder it's calendar plus notes.

Spend two minutes on Friday gathering. Drop everything into one doc. That doc is the input.

The three shapes

Same input, three different outputs depending on who reads it.

Status update for your manager. This is the 1:1 prep doc. Three sections: what shipped, what's blocked, what I need from you. Bullet points, not paragraphs. Your manager wants to know where to unblock you, not relive your week.

Project update for stakeholders. Progress against the goals you committed to, risks that have changed, decisions you need from the room. Stakeholders care about whether the thing is on track and whether you're flagging the right risks. They don't care about activity.

Board or exec summary. Short. Metric-led. Two or three numbers, two or three decisions, no list of things you did. The audit here is brutal: if it can't be skimmed in 60 seconds, it's the wrong shape.

The same week's artifacts produce three very different documents. Pick the shape based on the reader, not the writer.

The prompt

Here's the template. The few-shot pattern is doing the heavy lifting.

You're drafting my weekly report in my voice.

Below are two of my past reports that I'd be happy to send again.
Match the tone, the structure, the length, and the sentence rhythm.
Don't add headers I don't use. Don't add disclaimers.

[paste past report #1]
[paste past report #2]

Here is the raw material from this week:
- Calendar events: [paste]
- Linear tickets closed: [paste]
- Key Slack threads: [paste]
- Meeting transcripts: [paste]
- My scratch notes: [paste]

Audience: [my manager / project stakeholders / the board].

Shape: [what shipped / what's blocked / what I need]
       OR [progress / risks / decisions]
       OR [metrics / decisions, under 200 words]

Be specific. Use real names from the material. If something's
ambiguous in the source, flag it instead of inventing.

That's the whole prompt. Copy it. Save it as a snippet. Reuse it every Friday.

The voice trap

Most AI-drafted reports read like ChatGPT wrote them because the prompt didn't include voice samples. "Be concise" doesn't get you there. "Sound like me" doesn't either. What works is showing two real reports you sent in the last quarter that landed well.

The model has read more corporate prose than you'll see in a lifetime. Without samples, it averages all of it and gives you the beige middle. With two of your old reports pasted in, it suddenly sounds like you on a good day. Same trick from the marketer guide, the founder guide, and honestly every guide in this pillar. Show, don't describe.

Keep those two sample reports in a notes file. Reuse them every week.

The audit-it-once-a-month rule

After four weeks of AI-drafted reports, ask two questions. Did anyone notice? Did anyone act on them?

If no one noticed, you've validated the pipeline. The reports were good enough to be invisible, which is what reports are supposed to be.

If no one acted, that's the bigger signal. It's the one from guide one of this pillar: the report exists out of habit, not because anyone needs it. Stop sending it. Or replace it with a Slack message. Or replace it with a five-line email. The cost of the report should be lower than its value, and if no one's reading it, the value is zero.

The pipeline doesn't just save you time. It surfaces the work that shouldn't exist.

Wrapping the AI category

This is the last guide in the AI track in /learn. Quick map of where you've been.

Pillar 1 was the mental model. What this technology actually is, why it's different from the last wave, and which daily AI to settle into.

Pillar 2 was the role view. How sales reps, marketers, and solo founders are using it right now to do work they couldn't do alone before.

Pillar 3 was building with it. Vibe coding, the tool stack, and the path from a paragraph to a working prototype.

Pillar 4 was shipping it. Going from prototype to a product that real users trust, including the parts you can't hand to a model.

Pillar 5, this one, was about taking time back. Cutting the work that shouldn't exist, automating the work that should, and replacing the weekly report with a ten-minute pipeline.

The rest is up to you. Revisit any pillar when the situation changes. Share the workflow that actually saved you hours with the person on your team who needs it. The category isn't a curriculum. It's a reference, and it'll still be here next Friday.