Podcast Summary: How I AI – "How to create your own AI performance coach: Optimizing your unique nutrition, recovery, and injury management needs"

Guest: Lucas Werthein (Cactus)
Host: Claire Vo
Date: November 24, 2025

Episode Overview

This episode of How I AI explores how Lucas Werthein, head of technology at Cactus, built a bespoke AI-powered wellness coach using ChatGPT to synthesize and act on data from his personal health, nutrition, training, and recovery routines. Moving beyond “AI for business,” Lucas and Claire dive into granular, practical workflows that anyone can adopt to transform scattered health data into actionable, personalized advice, even for those not striving for six-pack abs or elite athletic performance.

Key Discussion Points & Insights

1. Genesis of the AI Coach

– Personal Problem, Personal Solution (00:00–03:55)

• Lucas, a lifelong athlete dealing with multiple injuries, found it increasingly difficult in his 40s to synthesize data from medical professionals, test results, and wellness routines.
  • Lucas: "Data is so siloed... The problem isn't the lack of data. The problem is the lack of synthesis and putting it all together." (03:55)
• After ChatGPT’s launch, he began feeding it his varied health data to create an actionable synthesis, leading to real breakthroughs in his own performance and recovery.

2. Types of Data Integrated into the Coach

– Multi-Modal, Multi-Format Inputs (10:13–13:20)

• Lucas’ GPT uses a wide array of data:
  • Imaging (MRIs, X-rays)
  • Activity & physiological cycles (CSV exports from wearable tech)
  • Daily journals (stress, recovery routines, anxiety)
  • Workout logs
  • Detailed sleep data
  • Serial bloodwork (PDFs in Portuguese and English)
  • Nutritionist-generated plans
  • Body composition scans
• Claire: "What I love about AI now... you can just dump all that data in. You don't have to worry about is it clean, is it organized, is it structured. Just put it in." (12:09)
• The system can process multilingual, unstructured, structured, and visual data, allowing for a comprehensive personal portrait.

3. Prompt Engineering for a Personal Coach

– Role, Goals, and Boundaries (13:20–17:49)

• The GPT is instructed to act as a “performance strategist and health optimization coach,” balancing real-world demands: an ambitious career, aging, and a desire to ‘feel 25 in a 40-year-old’s body.’
  • Key instructions:
    • Prioritize proven, accessible, scientific recommendations.
    • Current priorities: protect joints, amplify output, maintain pain-free movement.
    • Identify and flag red/yellow zones for overtraining, under-fueling, or inflammation.
  • Anti-prompts: Avoid pseudoscience, excessive supplements, or unmeasurable advice.
  • Emphasize actionable, context-sensitive recommendations.
• Lucas: “No fluff, no hacks, nothing that hasn’t been proven...” (13:20)
• Claire: “The idea of like a role and then a really realistic outcome for yourself is a nice framing..." (15:38)

4. Key Workflows: Nutrition, Recovery, and Injury Management

– Example Scenarios in Action (24:54–30:30)

A. Nutrition Planning for Real Life

• Adjusting diet proactively for social events (e.g., sake-heavy dinner).
  • GPT recommends clear strategies: high protein, low carb during the day before indulgence.
  • Feedback loop: Lucas sends a photo of his meal for instant feedback.
  • Claire: “You get like a very short blip of a good positive feedback loop... help people reinforce habits that compound over time.” (27:01)

B. Dynamic Adaptation & Accessibility

• The coach can respond to immediate life changes:
  • Unexpected parties, late nights, fast-food options, adjusting recommendations on the fly.
  • Strongly positioned as more accessible than human coaching, making basic optimization available to more people.
  • Lucas: “The fact that you now have a coach in your pocket is super interesting because things change.” (28:24)

C. Injury Management and Progress Tracking

• Inputting daily symptoms, images, and videos to track joint pain; validating medical advice; managing long-term stress around injury recovery.
  • Empowers user to ask if tournament dates are realistic, or if healing is on track.
  • "It is contextualized in a way that I can digest and now... the anxiety... is eased a little bit because I can manage the expectations..." – Lucas (30:30)

D. Synthesizing and Explaining Health Guidance

• GPT translates complex medical and PT advice into day-by-day actionable plans, new format outputs, or explanations in simpler terms (or in multiple languages).

5. The Meta-Framework: Cross-Expert Integration

– "The Ideal Team in One Tool" (20:15–21:28)

• Lucas’ prompt structure mimics how an ideal team (doctor, PT, nutritionist) would function if fully aligned—but in the real world, this is hard and rare.
  • Claire: "[This] reflects how in an ideal world, all these experts... would all be fully integrated, aligned... but that's not actually practical." (20:15)
• Brings disparate viewpoints and recommendations into a single strategy.

6. Prompt Tips for Listeners

– How to Structure Your Own AI Coach (21:48–23:43)

• Structure for success:
  • Assign a role and goal to the GPT.
  • Feed in real personal data and context.
  • Set clear anti-prompts and boundaries.
  • Specify output formatting, tone, and style.
  • Loop in values: precision, adaptation, context awareness.
  • Don’t hesitate to use ChatGPT itself to help structure prompts.
  • Lucas: "Write like a coach, be clear, tactical, no fluff, no lectures. Connect the dots..." (23:16)

Vision & Future of AI Health Coaching

– Synthesis at Scale (35:39–41:52)

• AI can be the “missing synthesizer” for personal health, especially given healthcare’s interoperability problems and siloed data.
• In 5 years, everyone could have a personal AI health coach—not to replace doctors but to make every micro-decision between visits more informed and contextually appropriate.
• Future: automatic, seamless data capture (smart fabrics, sensors, ambient devices), interoperable AI models trained on massive longitudinal population data.
  • Lucas: "The future of health isn't just about medical breakthroughs. It's... turning overwhelming data into something that's simple and very, very personalized." (37:51)
  • Potential for healthcare institutions to license proprietary “knowledge layers” for personalization.

Extending Use Cases: Not Just for Athletes

• Framework applies to caregiving, parenting, chronic disease management, general wellness, and even supporting children’s ambitions or elderly family members.
• Claire: "No matter what your goals are, setting up a framework like this... can help you day by day increment your way to them." (36:50)

Lightning Round & Additional Workflows (43:42–48:21)

• Lucas also builds synthetic clients (AI models replicating client styles for consulting work) and an AI “co-founder” as a business sounding board.
  • All synthetic models use only public information—no proprietary client data.
• Claire’s Tip: Build a synthetic version of yourself for your team to streamline workflows and feedback.

Notable Quotes & Memorable Moments

• Lucas Werthein (03:55): "The problem isn't the lack of data. The problem is the lack of synthesis and putting it all together."
• Claire Vo (12:09): “You can just dump all that data in... and then just figure out what comes out the other side.”
• Lucas Werthein (13:20): “No fluff, no hacks, nothing that hasn't been proven.”
• Claire Vo (27:01): “You get a short blip of a good positive feedback loop... that actually helps people reinforce habits that I think compound over time.”
• Lucas Werthein (35:39): “AI could be a missing synthesizer of personal health.”
• Lucas Werthein (37:51): “The future of health isn’t just about medical breakthroughs. It’s a lot about synthesis and the ability to turn this overwhelming amount of data into something that's simple and very, very personalized.”

Key Timestamps for Reference

• 00:00 – Lucas’s introduction and personal journey with health data
• 03:55 – Recognizing synthesis, not data, as the limiting factor
• 10:13–13:20 – Data formats used in the AI coach
• 13:20–17:49 – Prompt and configuration details
• 24:54–30:30 – Real world workflows: nutrition, recovery
• 30:30–35:39 – Using AI for injury management, visual/text inputs
• 35:39–41:52 – Future of AI in health, vision, and automatic data
• 43:42–48:21 – Additional workflows: synthetic clients, AI cofounders
• 48:49–51:08 – Final insights, learning from AI “mistakes,” and sign-off

Takeaways for Listeners

• Tailor your prompts: Be clear about your objectives, values, and hard lines.
• Use multi-modal inputs: Dump everything—images, videos, PDFs, notes—regardless of format or language.
• Iterate and adapt: Co-develop your GPT coach with your actual workflows and feedback.
• Broader applications: The framework is as useful for everyday wellness as for peak athletic optimization.
• AI is your synthesizer: Not a replacement for expertise, but a way to make sense of the fragmented modern health ecosystem—and bring coach-like support to all.

Guest links:
Find Lucas Werthein and his company Cactus at https://cactus.is/
Prompt template and show notes will be published alongside the episode.

End of Summary