Rep-by-Rep AI Personal Trainer

Sam is building an AI personal trainer that gives feedback after every set, based on your goals and plan. He took an existing version of the app into the new OpenAI Codex desktop Mac app to polish it, and demoed it in Episode 30 (February 6, 2026) at 08:01. He started building it the week before. It was still in beta: he hadn't used it in the gym yet.

Why we built it

Sam had been logging workouts in a ChatGPT thread, updating it after almost every exercise with his overall goal. It worked, but typing out each set was annoying. He wanted a simple interface and the kind of constant feedback a real trainer gives: you struggled on those 10 reps, today is a volume day, so drop the weight and hit your reps.

The stack

• OpenAI Codex desktop app with the GPT-5.2 Codex extra-high reasoning model: polished the existing app and found bugs.

Sam didn't say on the show which tool he used to build the first version or which model powers the in-app feedback.

How we built it, step by step

1. Start from what you have. Sam brought a previously existing version of the app into Codex.
2. Run extra-high reasoning to polish. The GPT-5.2 Codex model found "a ton of bugs" and gaps he hadn't noticed. Each run took five to eight, sometimes ten minutes.
3. Build the settings. Workout duration (his: 45 minutes), goal (get bigger vs. stronger), training days (minimum two; he chose five for a split), body focus (upper, balanced or lower, which shifts the mix to about 20% vs. 50%), experience level, equipment (full gym, home dumbbells, basic gym, body weight) and limitations.
4. Make logging one tap. Log a set and say how it felt.
5. Respond to effort. If you report a struggle but hit your reps with good form, it tells you to keep going. If you struggle and miss, a fatigue check steps you down. In the demo, weight dropped from 20 to 7.5 and the target reps came down too.
6. Check energy between exercises. Full tank, half empty or running low.

How it turned out

The demo worked end to end, but Sam called it "somewhat boring to look at" and said he hadn't used it much. His bigger takeaway was about the tools. He found Codex very similar to Claude Cowork in feel, and Claude Code more instantaneous and fun. Codex's long reasoning runs were slower and "less engaging," but good for the final polish. At 10:44 he put it this way:

using a slightly less reasoning model when you want quick iterations and then going to the deeper reasoning models when you want, like, you know, to be really begin polishing off and being thoughtful

— Sam Nadler, Episode 30

Jordan saw it as part of the personal-software trend: before, you would have used an app or a spreadsheet to track workouts; now it's easy to build exactly what you want.

What we'd do differently

• Match the model to the phase. Fast, lower-reasoning models for iteration; deep reasoning for polishing toward production.
• Test it where it will be used. We haven't reported real gym results yet, so treat this as a working prototype.

In the same episode, Jordan showed his personal DNA and health insights dashboard. For the tool question, see Claude Code vs OpenAI Codex. More on fitness: AI in Fitness.

FAQ

How do you build an AI personal trainer app?

Sam built a day-by-day workout app with settings for duration, goal, training days, body focus, experience, equipment and limitations. You log each set and how hard it felt, and the AI responds, for example stepping down the weight and reps after a fatigue check.

Can you build a workout app with OpenAI Codex?

Sam took an existing version of the app into the new Codex desktop Mac app and used the GPT-5.2 Codex extra-high reasoning model to find bugs and fill gaps. Each run took five to ten minutes.

Is Codex or Claude Code better for building an app like this?

Sam found Claude Code more instantaneous and more fun for quick iterations, and Codex's extra-high reasoning better for polishing toward production. His advice is to balance the two.

Is the AI personal trainer app finished?

No. It was still in beta at recording, and Sam planned to start testing it in the gym the following week.