Systems for Real Life

Let’s Connect

The AI Training Partner: How I used AI to build a smart, adaptive training plan without a coach.

04/13/2026

What happens when you give AI the right context and stay in the loop.

The Problem: After taking a step away from my training, I wanted to get back into a routine without hiring a running coach or building a plan from scratch.

The Result: I saved an estimated $150-$300, completed this in one hour and have a personalized plan to my exact body metrics.

The Process: How I built and refined my prompts and what I learned along the way.

Step 1: Building the Research Foundation.

Fig. 1: Beginning prompt.

I started by giving Claude detailed context about my body metrics, training frequency, and local climate. The output gave me a strong research foundation:

Fig. 2: Carbohydrate fueling rules.
Fig. 3: Sodium levels–daily and during training.
Fig. 4: Pre and post-run nutrition.
Fig. 5: Washington state context (avg 65ºF).

Step 2: Refining the Pace Data.

Fig. 6: Pace chart prompt.
Fig. 7: Pace chart output.

Step 3: Now that I had my research data complete, I needed to structure that data into something clear and easy to read.

Fig. 8: Structure of training schedule prompt.

The first attempted schedule was created:

Fig. 9: First schedule attempt output.

More iteration was needed on my part:

Fig. 10: Pace chart prompt.

I was getting close, but noticed some slight inconsistencies that made me pause so I prompted and asked:

Fig. 11: Prompting error question.

Claude’s response emphasizes why it’s vital to have a human in the loop. I caught an inconsistency and needed clarification as to why it happened.

Fig. 12: AI output error.

Before: Claude’s initial output (unedited).

Fig. 13: Before output.

After: I made edits within the spreadsheet so it looked cleaner and easier to read.

Fig. 14: After output.

The final prompt I entered was a vague 12 week half marathon training plan request.

Fig. 15: Vague prompt.

The results were straightforward but lacked any imagination.

Fig. 16: Vague training plan output.

The Output: The actual training schedule with pace chart and nutritional information specific to my age, weight, height, and where I’m currently at, not where my goals are.

Fig. 17: Final output.
Fig. 18: Final output cont.
Fig. 19: Weeks 6-7 zoomed in.

Biggest Takeaway: The structured prompting worked well to produce a plan personalized to my exact body metrics, the climate, fueling needs, etc. I adjusted the spreadsheet after pasting it into Google Sheets and caught and corrected an output error along the way. This entire case study saved me an estimated $150-$300 and only took one hour to complete.

The vague prompt produced a generic training outline that I feel many people and runners assume is all that generative AI is capable of.

Tools Used: Claude and Google Sheets

Skills: Iterative Prompting • AI Prompt Engineering • Data Interpretation • Systems Thinking • Human in the Loop  • Generative AI Application