I Log Every 5 Minutes of My Day (And Let AI Roast Me)

I Log Every 5 Minutes of My Day (And Let AI Roast Me)



Tópico: I Log Every 5 Minutes of My Day (And Let AI Roast Me)
Categoria: Tutoriais | Programação & Tecnologia
Idioma Principal: Português (Conteúdo de Tecnologia)

Descrição do Conteúdo / Informações:
-------------------------------------------------------------------------


Gus Fring had the right idea


You know the scene in Breaking Bad where Gus gets interrogated by the cops? They ask him where he was. He pulls out a little book with every hour and every place he'd been written down. The cops can't touch him because his alibi is flawless.

I wanted my own version of that little book.

It started on a family trip, a staycation at a villa where I slept the entire time. Like, aggressively slept. Literally, my body somehow shuts down whenever I touch a soft surface, and boy there was a lot of soft surface. And when I got home I was still exhausted. I couldn't figure out why I was so tired all the time even though I was supposedly resting. So I started logging. Somehow I've kept it going every day for ten months now. Here's what a typical Friday looks like:

📅 Daily Summary – Fri, 12 Sept 2025
⏱️ Total Logged: 15h 50m

🧱 By Category:
Category          Time     %
─────────────────────────────
Focused Work    6h 20m   40.0%
Distraction     3h 20m   21.1%
Personal        2h       12.6%
Family          1h 25m    9.0%
Meeting         1h 5m     6.9%
Exploration     1h        6.3%
Admin Work      40m       4.2%
Planning        20m       2.1%

🔁 By Type:
Type              Time     %
─────────────────────────────
Work            9h 25m   59.5%
Wasted          3h 20m   21.1%
Recovery        2h       12.6%
Social          1h 25m    9.0%

With this log I can immediately see that 21% of my logged day went to doom scrolling social media for no good reason. I wouldn't have known that if I hadn't written it down.



Why 5-minute blocks


The original goal wasn't even time tracking, it was mood tracking. I wanted to know whether a morning workout changed how I felt for the rest of the day versus skipping it. The log was the vehicle to find out.

I tried screen time reports first. There was no clean way to pull that data into Notion, and copying it across by hand every day takes too much effort. Screen time has a deeper flaw anyway: it tracks which app you opened and never what you were doing inside it. Chrome can't tell whether I was debugging something or rereading the same Reddit thread for the third time. For working out where your attention actually goes, app-level data is close to useless.

Then I played with granularity. Logging to the minute was brutal. I'd agonize over whether something started at 10:07 or 10:09, and within a week the whole thing felt like a chore I wanted to quit. I tried 15- and 30-minute blocks too, but they ran way too loose. You can bury a lot of doom scrolling inside a 30-minute block and still tell yourself you were honest about it.

After all that, five minutes turned out to be the sweet spot. It's detailed enough that I couldn't mess it up, but still loose enough that I never dreaded keeping it up.

Two sets of labels:

• Categories (what you're doing): Focused Work, Distraction, Exploration, Admin Work, Personal, Meeting, Family, Planning.

• Types (how to count it): Work, Wasted, Recovery, Social.

The two layers pull apart "what did I do" from "was it worth it". Personal time, playing with my kid, recovery, none of that is wasted. Doom scrolling is.



How the system evolved




Phase 1: n8n (July 2025)


I logged each activity in Notion with a start time, end time, description, and category. Then I built an n8n workflow that pulled the day's entries, expanded each one into 5-minute blocks, and pushed a summary to Telegram every morning.

The script handled midnight crossovers, meaning anything that ran past 12am, snapped timestamps to 5-minute boundaries, and worked out the category and type breakdown with percentages. It took a few passes to get right, and the midnight logic alone took three rounds of debugging.

n8n worked. But it cost me $20 a month for the cloud service and was one more thing in the stack that could break. It was the automation tool of the moment though, and everyone was on it.



Phase 2: OpenClaw (early 2026)


OpenClaw blew up in January 2026 and it beat n8n for this. One tool read my Notion entries, called Haiku to categorize them, and pushed the summary, so I moved everything over.

Then Anthropic started charging extra for third-party agents routing through Claude. OpenClaw was still solid, but now it cost real money to run, so I started looking for something else.

Phase 3: Claude Cowork (2026 to now)

This is what I use now. I set up a scheduled task in Claude Cowork that reads my Notion log every morning and sends me the same daily breakdown n8n used to, except now there's no server to babysit and nothing to maintain. When I want more than the morning summary, I ask for it. I can pull up my distraction pattern for the week, or check whether I'm in more meetings than I was last month, and it answers off the same data.

It also bailed me out of a question I used to be terrible at. When my wife asks what we did at some restaurant back in March, I spin up Claude, which is already wired to my Notion over MCP, and it gets the answer right every time. Everything's searchable, trips, restaurants, outings, anything.



What I learned


The migration was never a hunt for the simplest setup. Each tool was the right call on the day I picked it. n8n won in July 2025, OpenClaw won when it launched, and Cowork won once the ecosystem moved. Every tool got replaced eventually. The logging never did.



What never changed


Across all three phases, the input stayed manual. I open Notion and type the time and what I'm doing every time I switch tasks. No app tracking or automatic detection of any kind.

This is what I'm proud of. Ten months without a single skipped day. The accuracy slipped when I was traveling and the entries got vaguer, but I still logged something every single day. I'd take the unbroken streak over a perfect one.



What the data actually showed


After a few months of logging, patterns emerged that I couldn't see when I was just "feeling" productive.

Distraction clusters after lunch. The 2pm to 4pm stretch was my weakest almost every single day. The cause was the dead air between tasks: I'd reach for my phone to fill it and lose 20 to 40 minutes before I even noticed.

Meetings fragment the day worse than I expected. A 30-minute meeting doesn't cost 30 minutes. It costs the 15 minutes before it, when I can't start anything deep, plus the 10 after, while I claw my way back into focus. Three 30-minute calls on a "light" day quietly burn two and a half hours.

The push-crash cycle took weeks of data to show up. My most productive days were also my latest nights. I'd push until 1 or 2am, wake up groggy, scroll away the morning, feel behind, then push late again to make up for it, the same loop running over and over. Those late nights weren't productive at all. I was borrowing against the next day. Living inside the loop I never noticed it, but the log laid it out flat in front of me.



The AI feedback loop


The payoff comes from handing the day to an AI afterward and asking what it notices. Give it your breakdown and it'll surface things you already half-knew. "You doom scrolled 90 minutes between 2pm and 4pm on four of five weekdays. That's six hours a week." You knew that already. But seeing it counted out in a flat sentence with no emotion behind it lands differently from carrying it around as a vague background feeling.

The AI never lectures me about discipline. It hands back the numbers, and a number is harder to argue with than a feeling.

Almost a year in, that September Friday's 21% sits at 10 to 12% on most days. Willpower had nothing to do with it. Once you know you'll have to type "doom scroll" into the log and then watch it swell your daily total, the scroll stops feeling free.



How to start


You don't need n8n or OpenClaw or any automation at all.

Open a note. Every time you switch tasks, jot down the time and what you're doing. At the end of the day, tag each entry and total it up, and that's it.

Categories I'd start with:

Focused Work (deep, uninterrupted, meaningful)

Shallow Work (emails, messages, admin, necessary but not deep)

Distraction (doom scrolling, aimless browsing, phone-picking)

Recovery (meals, rest, exercise, deliberate breaks)

Social (family, friends, calls that aren't meetings)

Don't automate until you've done it by hand for two weeks. The awareness comes from the manual logging, and automation only keeps it alive once the habit is already there.

Once you've got two weeks down, point an AI at it. I'd log in Notion since the MCP support is solid and Claude can query it directly without any exporting or copy-paste. Ask what patterns it sees. The honest answer is usually the one you've been avoiding.

Making this into an app has been on my mind, something that makes the logging fast, the analysis automatic, and the AI feedback loop built in. If that sounds like something you'd use, drop a comment or DM me.