Ben Newton - Commerce Frontend Specialist

I Track Every Workout By Talking To My Phone. Here Is The System.

I Track Every Workout By Talking To My Phone. Here Is The System.

I was sitting on the upright chest press machine, phone in hand, sweat on my forehead, typing into a Claude chat.

"I just did 3 sets of 10 at 60 pounds. Log it."

That's it. No app. No spreadsheet. No fiddling with a stopwatch and a notes app. I closed the chat, racked the weight, and moved to the next machine.

Two minutes later I asked Claude, "what did I do last time on chest press?"

It told me. Date, weight, sets, reps. Then it told me the time before that. Then it asked if I wanted to move the pin up a notch this week.

This is not an app I downloaded. This is a system I built. And before anyone fires up the "this is just an ad disguised as an article" complaint, like a couple of folks did on my last post, let me save you the trouble: yes. I write software. It says so in my bio. I am building this product, I am using it every day, and I am writing about how I use it. If that's a problem for you, the unfollow button is right there.

For everyone else who wants to know how the gym workflow actually works, keep reading.

Why I needed this in the first place

I am 55. I was overweight for years. Over the last year I lost a lot of it. A month ago I walked into a gym for the first time in many many years as someone who was actually going to lift weights, not just look at them.

I have no baseline. No idea what I can lift. No idea how many reps is the right number. No idea what "good form" feels like on a machine I have never touched. I am genuinely starting from zero at an age when most lifters have thirty years of muscle memory to draw on.

I am also not doing free weights. I am on machines. Pin-loaded stacks, named exercises, fixed motion paths. That is a deliberate choice for where I am right now. The machines remove a class of mistakes I do not need to be making at month one.

And here is the thing nobody tells you about being a beginner at 55: the last thing in the world I am going to remember, three days later, is what I set the pin at on the triceps machine. It is not top of mind. It will never be top of mind. I am thinking about not falling, not pulling something, not embarrassing myself. The number on the stack is not getting filed away in my brain for later retrieval.

But without that number, there is no progression. There is no "move the pin up a notch this week." There is no knowing if I am getting stronger or just showing up. The whole game of building a body in a gym depends on remembering what you did last time, and I cannot remember.

So I built a memory I do not have to maintain. A few weeks in, the log is real. The progress is real. I would not have either of those things without this notetaking routine. That is what this post is actually about.

The setup, and what BlackOps actually is

Here is the part I want to be clear about. I did not build BlackOps Center for fitness logging. I built it to connect Claude and ChatGPT to my notes, originally so I could turn ideas into published content without losing them in fifteen years of half-finished drafts across Apple Notes and Obsidian. It is a writing and publishing system at its core.

But like the truck story I told last week, the same shape that works for one thing turns out to work for a lot of things once you start using it. Connect an AI to your notes. Give it tools to read and write them. Wire in whatever external signals are already being captured for free. The pattern is generic. The gym workflow is one application of it.

So here is the setup. I have a personal BlackOps Center workspace synced to an Obsidian vault. Inside that vault is a folder called fitness. Inside that folder is one Markdown file per exercise. Chest press. Triceps press. Shoulder press. Lat pulldown. Preacher curl. Ab crunch. Cardio treadmill. They are dumb little tables. Date, sets, reps, weight, notes.

That folder is registered as a brain in BlackOps. A brain is a recursive compile of every note in a folder into one queryable context note, served by the get_brain MCP tool. The compile runs automatically. I never touch it. When I ask Claude what I did last time, it is calling get_brain on the fitness folder and reading the compiled memory back.

When I open Claude on my phone, I have a skill installed called fitness-logger. A skill, in Claude Code terms, is a folder with a SKILL.md file that tells the model how to behave when a particular kind of request shows up. Mine knows three things:

  1. How to write a new row into the right exercise file when I report a set. It does this by calling post_notes against my BlackOps workspace, which writes the Markdown and syncs it down to Obsidian automatically.
  2. How to read the brain when I ask what I did last time. That is get_brain on the fitness folder.
  3. How to pull my Apple Watch data from HealthKit and add it to the session.

That's it. That's the whole product surface. Three behaviors, one folder, one phone. The skill orchestrates the BlackOps MCP tools. The MCP tools handle the writes and the compile. Obsidian gets the synced copy for free. None of this was built for fitness. All of it works for fitness.

The phone in the gym

Here is what an actual session looks like. I finish three sets on the upright chest press, drop the pin back into the rest position, and stay sitting on the machine. I open the Claude app on my phone. I tap the dictation button and say:

"Did 3 sets of 10 at 60 pounds on the upright chest press."

The skill fires. fitness-logger parses the dictation, finds chest-press.md, and calls post_notes with an appended row: today's date, 3 sets, 10 reps, 60 pounds. The note is written. The Obsidian sync picks it up. The brain marks itself stale for the next compile. Total elapsed time, three seconds. By the time I am wiping the machine down, the confirmation is on the screen.

I move to the triceps press. I sit down on the machine and before I touch the pin, I ask Claude what I did last time. It answers instantly:

What did I do last time on triceps?
Last session was 3 sets of 10 at 50 lbs. The session before that was also 3 sets of 10 at 50, and the note said it felt easy. Want me to set you up for 55 this time?

That is fitness-logger calling get_brain on the fitness folder and reading the compiled memory back. I do not have to remember anything. The system remembers for me.

I set the pin at 55, do my three sets, sit there afterward, and dictate the row. "3 sets of 10 at 55 on triceps." Same write flow as before.

Then I walk over to a machine I have never used before. This is the part I want to slow down on, because this is where being a 55-year-old beginner actually matters. I do not know what this machine is called. There is a padded armrest, an angled bench, and a bar hanging in front of it. I do not want to guess and end up logging three months of work to the wrong file. So I take a picture of the instructional graphic on the side of the machine. The one with the name of the exercise, the muscle group diagram, and the little stick figure showing you how to use it. I send it to Claude with no other words.

The skill reads the graphic, identifies the exercise as "preacher curl," creates preacher-curl.md in the fitness folder on its own with the right headers, and tells me it is ready for my first set. I did not have to ask it to make the file. I did not have to confirm the name. The skill knew there was no existing note for this exercise and made one. From that point forward, every "did 2 sets on the preacher curl" gets logged to the right place. The picture did the disambiguation. The skill did the bookkeeping. I just used the machine.

Then shoulder press. Same dictation flow as the others.

Last station, treadmill. I do a 30 minute block. When the cooldown is done and the belt stops, I do not dictate the speed and incline. I take a picture of the dashboard. The display with the time, distance, calories, and incline lit up in red LEDs. I send it to Claude with no caption. The skill reads the photo, extracts the numbers, pulls my heart rate data from the Apple Watch for the same window, and calls post_notes to write the full row into cardio-treadmill.md.

By the time I walk out to the parking lot, every set is logged, every cardio block is logged, and the brain knows about all of it.

Here is what that logging exchange actually looks like in real time.

Now here is the part that flips this from "voice note app" to "system."

The Apple Watch closes the loop

When I ask Claude to log the cardio session, it does not just take my word for it. The skill is wired to read my HealthKit data. It pulls the actual workout the Watch recorded. Average heart rate. Peak heart rate. Calories. Distance. Time spent in each heart rate zone.

It writes those numbers into the note alongside what I told it via post_notes. So the cardio row I dictated as "30 minutes, 2.7, 3.5" gets enriched with my average heart rate, my peak, and the time I spent in each zone.

I did not type any of that. I did not open the Health app. I did not even take the Watch off my wrist. Claude went and got it.

This is the part that took me a minute to internalize. I am wearing a sensor that is logging real biometric data in the background. That data has always been there. What was missing was a way to attach it to the thing I was actually doing without breaking my focus to do data entry. The skill is the bridge. The brain is the memory. The Watch is the witness.

I just take a picture of the dashboard

It is worth restating, because it is the most distinctive thing about how this actually works in practice.

I do not dictate the speed and incline anymore. I do not log a new machine by name. I point my phone at whatever piece of equipment I am standing in front of, and the right thing happens. The treadmill dashboard turns into a row of cardio data. An instructional graphic on a machine I have never used turns into a new exercise file. A weight stack turns into a logged set.

I did not type anything. I did not dictate anything. I pointed my phone at a piece of gym equipment and the row appeared in the file. That is what "the disappearance of the act of taking notes" actually looks like in practice. The system is meeting me where I already am, instead of asking me to translate what I just did into structured data.

What the folder actually looks like

Here is the fitness folder in my vault right now:

fitness/
├── ab-crunch.md
├── cardio-treadmill.md
├── chest-press.md
├── lat-pulldown.md
├── preacher-curl.md
├── shoulder-press.md
└── triceps-press.md

Each file is a Markdown table. No schema, no app, no database. The get_brain compile reads all of them and stitches them together into one queryable context.

Here is a real row from cardio-treadmill.md:

DateDurationSpeedInclineCaloriesAvg HRMax HRNotes
2026-05-2030 min2.7 mph3.5177118130Distance 1.39 mi

Some of those values came from a photo of the dashboard. The rest came from the Watch. I did not type any of them, and I did not open the Health app once.

Here is the full chest-press.md table as it stands today, pulled straight from the file:

DateSets×RepsWeightNotes
May 64 × 1035 lbs
May 113 × 1060 lbsBackfilled from workout note
May 213 × 10/10/960 lbsSet 3 stopped at 9. Ten-day gap flagged by the brain.

Three rows. Two and a half weeks. I jumped from 35 to 60 between the first and second session. I missed reps on the third. The system logged it anyway. The system also flagged the ten-day gap, because the brain compile noticed I had no entries between May 11 and May 21. That is the thing I cannot get from a notes app or a spreadsheet. The memory is doing work I did not ask it to do.

Here is the full triceps-press.md table:

DateSets×RepsWeightAvg HRNotes
May 123 × 1050 lbs-First logged session
May 183 × 1050 lbs118Repeat of last session
May 213 × 1055 lbs ↑95 ↓Bumped from 50. HR dropped, work is getting easier.

That is the entire progression on one exercise over nine days. Without the log, I would have no idea I had moved from "first logged session at 50" to "bumped weight from 50" in nine days. I would also have no idea my average heart rate dropped from 118 to 95 between sessions at the same weight, which is a real signal that the work is getting easier. With the log, the next session has a built-in instruction: try 60. That instruction lives in the notes column because I told the skill to write it there. Next time I ask the brain what to do on triceps, it sees it.

This is progression. This is what beginners are supposed to do and almost never actually do, because they cannot remember what they lifted last time. I can, because I am not the one remembering.

And here is what asking about your last workout looks like once the data is in.

Why this matters more than gym tracking

Forget the gym for a second. The pattern here is the point.

A brain is a folder of related notes that compiles itself into a queryable memory, accessed via the get_brain MCP tool. A skill is a small set of instructions that teaches Claude how to read and write that memory through tools like post_notes. You wire one or two external data sources into the skill, and now you have a system that takes voice input, enriches it with sensor data, writes structured records, and answers questions about its own history.

I am using that pattern for workouts because workouts are simple and the feedback loop is fast. But the same shape works for anything you want to track over time without becoming a full time data entry clerk.

Your sales pipeline. Your client meetings. Your blog post performance. Your kid's reading log. Pick a thing you wish you had a memory of, build a folder, write a skill, point it at whatever external signal is already being captured for free.

This is what BlackOps Center actually is. I built it to write and publish. I did not build it to track workouts. But the same plumbing that connects Claude and ChatGPT to my notes for one purpose works for any other purpose, because the plumbing is the product. The truck post I wrote last week was the first time I named it in public. This post is the second time. I am going to keep naming it because the whole point of building a product is that you actually ship it.

What I am not telling you

I am not selling you a 12 week transformation. I am not telling you I dropped twenty pounds in the gym, because I dropped most of it before I ever walked in. I am not telling you my one rep max because I do not have one yet, and at 55 I might never need one. What I am telling you is that for the first time in many many years, I know exactly what I did last Tuesday on shoulder press. That is a baseline. Next week it will not be.

The system works because it gets out of my way. The phone goes in my pocket. The Watch keeps watching. The brain keeps compiling. The skill keeps writing.

And when I want to know what I did, I ask.

That's the future of note taking. It is not a better notes app. It is not a smarter editor. It is the disappearance of the act of taking notes, because the system is already doing it for you.


BlackOps Center is the infrastructure. I built it. I use it every day. You can too: blackopscenter.com.

I wrote this post inside BlackOps, my content operating system for thinking, drafting, and refining ideas — with AI assistance.

If you want the behind-the-scenes updates and weekly insights, subscribe to the newsletter.

Related Posts