my2sats
← Back to posts
Coached by if-checks - OpenClaw Personal Training Day 1

Coached by if-checks - OpenClaw Personal Training Day 1

egge·
openclawaifitnessagentcoach

For most of the past decade, I haven’t been particularly fit. The last time I had single-digit body fat was right after high school. It never really bothered me. But now, as a parent, things feel different.

I want to lead by example and at the same time stay as healthy as possible so I can be there for my child. Additionally I signed up to become a voluntary fire fighter and running out of breath in a burning building is not an option.

So I’ve set a clear goal for the next three months:

  • Lose 10 kg
  • Improve my 5K time from 31:00 to 24:30

And yes, you might reasonably ask:

“Why are you writing about this on your technical blog?”

Because I’m not doing this alone. Meet my new personal trainer: Coach Claw.

My Digital Personal Trainer: OpenClaw

I’ve written about OpenClaw before. In short, it’s an AI agent on steroids.

By giving a regular AI agent both a heartbeat (.md) and a soul (.md), it evolves into something much more practical: a persistent personal assistant that retains memory, understands context, and interacts like an actual PA.

Over the past few weeks, I’ve experimented with a lot of capabilities. But this time, we’re pushing things further. This time, OpenClaw becomes my coach.

Briefing Coach Claw

I kicked off the journey by clearly stating my objectives:

  • Lose weight
  • Improve my running
  • Become operationally fit for firefighter duty

Coach Claw responded exactly how a good trainer should: with questions. He sent me a questionnaire to assess my current level, history, and constraints. Once that baseline was established, he prescribed a 5K benchmark run.

It went… poorly.

After only weeks of running again, my pacing instincts were nonexistent. I went out too hard and burned out fast, stopping at 3 km in 16:30.

Not glorious, but still useful. And that’s what mattered. Coach Claw had enough data to generate a structured four-week training block, written cleanly into markdown files.

Here’s an excerpt from Week 1:

Week 1 Training Plan

  • Mon — Run (Easy):
    40 min easy + 4 × 20s relaxed strides (full walk-back recovery)

  • Tue — Strength A (Push + Pull + Core):

    • Bench press 5×8 (moderate, 2 reps in reserve)
    • Row variation 4×10
    • Push-ups 3 sets near technical failure
    • Plank 3×45–60s

  • Wed — Run (Intervals):
    10 min warmup
    6 × 2 min hard / 2 min easy jog
    10 min cooldown

  • Thu — Strength B (Legs + Core):

    • Squat pattern 4×8
    • RDL / hip hinge 3×10
    • Walking lunge 3×10 each leg
    • Side plank 3×30–45s each side

  • Fri — Run (Easy Long):
    55–60 min easy

  • Sat — Strength C (Bench endurance focus):

    • Bench @ 40 kg: 5 sets submax (stop with 2–3 reps in reserve)
    • Incline push-up or dip 3×12–15
    • Shoulder press 3×10
    • 8–10 min core circuit

  • Sun — Run (Steady Tempo):
    10 min warmup
    3 × 8 min steady-tempo (6:20–6:40/km) with 3 min easy jog between
    10 min cooldown

How Much Guidance Was Required?

Quite a bit, but intentionally. Having experimented with OpenClaw before, I know that AI agents perform best with explicit structure. So I gave very concrete instructions:

  • Write everything to disk
  • Keep baselines and benchmarks persistent
  • Use markdown files to structure and track workouts
  • Log progress consistently
  • Read the workout directory before discussing training

I also explicitly tell Coach Claw when to write to memory.

In other words, I’m not just training my body I’m also training the agent.

What’s Next?

There’s still plenty to build:

  • Automated progress tracking
  • Scheduled weigh-ins
  • Benchmark reminders
  • Cron jobs for daily workout digests
  • Performance dashboards

But one step at a time.

For now, I run. I lift. I log. And I let an AI yell at me when I skip leg day.