Week 40: Using AI as my gym coach

The Experiment

I've been going to the gym twice a week for 10 years. I'm 6'2", I run 20km weekly, I swim, I surf, I do yoga. By most measures, I'm active and consistent.

I'm also weak as hell.

After a decade of showing up, I was still squatting 20-25kg. The same exercises, the same weights, the same total lack of progression. I had no idea what I was doing, no plan, and no visible results to show for 10 years of effort.

Then Peep Laja mentioned he was using ChatGPT as a gym coach. I thought: why not test it?

The Process

I chose Claude for this experiment (mostly because I'd been using it for other projects and the conversation interface felt natural).

1. Starting with data collection

Week 1, Claude just watched. I did my normal workout and logged everything in real-time:

Claude asked questions after each session:

• How did the form feel?

• Which exercises felt hardest?

• Any discomfort or pain?

2. The brutal assessment

Week 2, Claude had seen enough. The diagnosis: I was massively underloading squats.

We did a progressive test: bodyweight squats, then 40kg, 50kg, 60kg. I hit 60kg for 8 reps with what Claude rated as 7/10 form.

I'd jumped from 25kg to 60kg in one session. Not because I'd suddenly gotten stronger, but because I'd been training at 40% of my actual capacity for years.

3. The programming

Claude set up a push/pull/legs split across my 3 weekly sessions:

• Session 1: Upper push + squats (bench press, overhead press, squats)

• Session 2: Upper pull (pull-ups, rows, lat pulldowns)

• Session 3: Deadlifts + full legs (deadlifts, squats, accessories)

The rule was simple: when I completed all sets with good form (8/10+), add 5kg next session.

4. Real-time logging

During workouts, I'd message Claude:

Me: "Squats 3×8 at 65kg, form felt good, depth was better"

Claude: "Squats: 3×8 at 65kg ✓
Great progression! That's +5kg from last week's 60kg. 
Form improvements noted. Consistent depth is exactly 
what we want to see.

Next session target: 70kg for 3×8 if today's form 
was 8/10+. Ready to add lateral raises?"

Immediate feedback, tracking, and next-session planning all in the conversation.

5. The setback

Week 4, I tweaked my lower back on deadlifts. Slight sacrum soreness that wouldn't go away.

I told Claude. It immediately adjusted the program:

• Skip deadlifts entirely

• Reduce squat loading to maintenance weight

• Focus on upper body while monitoring recovery

• Provide specific form cues for when I returned to deadlifts

This is where Claude wanted me to film my deadlifts and upload them for form analysis. I didn't (embarassment, mostly). Instead, I asked for best practice form cues and worked through it more cautiously.

The Outcome

Six weeks of AI coaching versus 10 years of winging it:

Squat: 25kg → 70kg (+180%) Deadlift: 60kg → 70kg (+16.7%) Bench Press: 40kg → 40kg (+0kg, but form improved significantly) Overhead Press: 32.5kg → 37.5kg (+15.4%) Pull-ups: 12 total reps → 18 total reps (+50%)

The squat progression is absurd because I was starting from such chronic underloading. But even the "slower" progressions (deadlifts, overhead press) represent more change in 6 weeks than I'd seen in years.

Beyond the numbers: I now understand what progressive overload actually means. I know when to add weight, when to work on form, and when to back off. Claude taught me to think like someone who knows what they're doing.

What worked:

• Real-time logging meant immediate feedback and motivation

• Having someone (something?) track everything removed all mental load

• The conversational format felt natural, easier than following a static program

• Claude adapted when I reported the back issue, didn't just blindly push forward

What didn't:

• I should have filmed my deadlifts when Claude suggested it

• The back tweak could have been avoided with better form checking

• Some exercises (bench press) need more than just "add weight", they need technical work

Key Takeaway

Ten years of gym experience means nothing if you're not applying progressive overload systematically. AI coaching forced me to track, measure, and progress in ways I'd never bothered with before.

Is it perfect? No. I've got a niggling back injury from not following Claude's advice properly. But in 6 weeks I've made more measurable progress than in the previous decade.

Pro Tips for Beginners

1. Start with data collection: Don't change anything Week 1, just log everything you currently do. Let the AI see your baseline.

2. Be honest about form quality: Rate your form 1-10 for every exercise. If you're below 7/10, don't add weight.

3. Log in real-time: Don't try to remember your workout later. Message the AI between sets.

4. Follow the advice you're given: When Claude says "film your deadlifts," don't be lazy like me. Form matters more than weight progression.

5. Split your training: Push/pull/legs works well for 3× weekly. Gives each movement pattern recovery time.

Want to Try It Yourself?

• Use Claude (or ChatGPT) in a new Project

• Add in context (I added some Hubermann documents, but there is plenty of gym advice out there)

• Ask it to coach you through progressive strength training

• Log every workout in real-time with weights, sets, reps

• Report how exercises felt, form quality, any discomfort

• Actually film your form when it asks (learn from my mistake)

I'm only 6 weeks in. I'll report back at 12 weeks to see if the progression continues or if I've plateaued. Either way, it's already been more useful than a decade of guessing.