Stop prompting. Start observing.

Stop prompting. Start observing.

Before you touch a single AI tool, spend a week watching how you actually work. Where do things take longer than they should? What do you redo every Monday? Where do you copy-paste between three tabs to produce one paragraph? Where do you make the same judgment call over and over?

That's your system. It already exists. It's just manual.

AI doesn't replace that system. It runs it. But first you have to see it clearly enough to describe it — what goes in, what comes out, what "good enough" actually means at each step. The work happens before the prompt.

The wrong question

Most people approach AI with the wrong question. They ask "what can it do?" and end up with a tab full of demos and zero time saved. The better question is quieter and a little uncomfortable: what am I doing every week that I shouldn't be?

You can't answer that from memory. Memory flatters you. It remembers the work you're proud of and forgets the 40 minutes you spent reformatting a spreadsheet on Tuesday.

You need data. Real, ugly, in-the-moment data.

The one-week audit

The protocol is simple enough to start tomorrow.

For five working days, every time you begin a task, jot it down. Paper, Notes app, the back of an envelope — whatever you'll actually keep open. Title and rough duration. That's it. Don't categorize. Don't judge. Don't try to remember at the end of the day. Capture in real time or you'll lie to yourself.

The list should feel ugly by Friday. "Reformatted slide deck, 25 min." "Replied to vendor about invoice, 10 min." "Rewrote Slack message three times, 15 min." Good. That's the data.

Friday afternoon, sit down and sort the list into four buckets. Most people are shocked by what shows up where.

The four buckets

Keep manual. The work where being human is the point. Customer relationships. Hard hiring calls. Original thinking. The writing where your taste is the product. If a model could do this as well as you, your edge wasn't taste — and you have a bigger problem than your calendar.

Eliminate. The standup that could be a Slack thread. The weekly report nobody opens. The status update that exists because somebody three years ago didn't trust someone who no longer works there. The recurring meeting you inherited. Killing these beats automating them, every time.

Hand to a person. Cleaner-than-you-do-it work that isn't your highest leverage. Inbox triage. Travel booking. Light research. Basic bookkeeping. A VA, a freelancer, or a teammate who actually likes this work will do it better and faster than you.

Hand to AI. Research briefs, draft writing, transcription, summarization, data wrangling, repetitive lookups, formatting, light coding. The tasks where the prospect, reader, or customer doesn't care that a human did it. That last line is the test. If a human's fingerprints don't change the outcome, the task belongs in this bucket.

The "automate this" pile is usually smaller than people expect. The "why am I doing this at all" pile is usually larger.

The automate-it-anyway trap

The most common mistake is automating work that shouldn't exist.

If a report takes you an hour a week and you build an AI workflow that produces it in two minutes, you feel like a hero. But if nobody actually reads the report, you've just built a faster way to waste two minutes a week — plus the hour you spent setting it up.

Killing the report is always cheaper than automating the report.

Before anything moves into the AI bucket, ask: would the world notice if this stopped happening? If the honest answer is no, eliminate. Don't optimize.

Stop chasing other people's systems

Right now everyone is chasing someone else's system. "Comment 'Claude' below." "I'll DM you the system." Cool. But that solved their problem, not yours.

The reason these templates feel hollow when you try them is that they were tuned to a different week. Different stakeholders. Different judgment calls. Different "good enough" bars. The system shape only works for the work shape it was built around.

Stop trying to crack someone else's code. Ask yourself the harder question: what's the most manual, repetitive part of my own workflow right now? Once you can describe one of those clearly — what triggers it, what inputs it needs, what an acceptable output looks like — you've done the actual hard part. The prompt is the easy part. The MCP server is the easy part. The agent is the easy part.

A worked example

A real-ish product manager's Friday-afternoon sort:

• 1:1 with my report (45 min) → Keep manual
• Wrote weekly update for leadership (60 min) → Hand to AI (summarize from notes + Slack)
• Reformatted competitor screenshots into slides (40 min) → Hand to a person
• Recurring "ops sync" meeting (30 min) → Eliminate (could be a doc)
• Drafted PRD for new feature (90 min) → Keep manual
• Hunted through Linear for tickets to mention in the update (25 min) → Hand to AI
• Replied to 14 internal questions in Slack (50 min, scattered) → Keep manual (mostly)
• Wrote three versions of a tricky message to a frustrated customer (35 min) → Keep manual
• Pulled retention numbers from three dashboards into one sheet (40 min) → Hand to AI or eliminate (does anyone use the combined sheet?)
• "Quick" status email to a stakeholder who never replies (15 min) → Eliminate

Two real wins for AI. Two clean eliminations. One handoff. The rest stays.

The total time saved is bigger than the AI bucket alone, because the eliminations save you time forever, not just this week.

The order matters

Run the audit before you build anything. Be honest on Friday. Resist the urge to automate everything in the AI bucket on Monday morning — start with one item, the most painful and most repetitive, and get it actually working before adding a second.

For most people, the first item that should move is email. High frequency, low signal, patterns that repeat. Start there. Get one thing right before adding the next.

The promise of AI isn't that it gives you a faster way to do everything you currently do. It's that it lets you see, finally, how much of what you currently do isn't worth doing.