How I AIHow to create your own AI performance coach: Optimizing your nutrition, recovery & injury management
CHAPTERS
- 0:00 – 4:55
Lucas’s athletic drive, surgeries, and the search for a ‘25-year-old’ body at 40
Lucas explains his lifelong intensity with sports, the accumulation of injuries and surgeries, and how turning 40 changed the stakes of recovery. He frames the core problem as not lacking data, but lacking a way to synthesize it into clear daily actions.
- •Competitive sports background (surfing, Muay Thai, tennis, weightlifting) and repeated injuries
- •Multiple surgeries (knees, shoulder, foot) and ongoing joint protection needs
- •Desire to maintain high performance while running a company
- •Health information overload: many experts and tests, little synthesis
- •Motivation to use ChatGPT as a personal “performance strategist”
- 4:55 – 6:11
Why health optimization breaks down: siloed experts, missing links, and poor coordination
The conversation highlights how medical and performance advice often fails due to fragmented ownership—doctors diagnose, PTs rehab, but nobody integrates biomechanics, training load, nutrition, and daily habits. Lucas positions AI as a coordinator that can connect these dots consistently.
- •Common gap between diagnosis, treatment, and root-cause mechanics (e.g., stroke/grip/biomechanics)
- •Disparate inputs: nutritionist, PT, wearable metrics, imaging, labs, journaling
- •AI’s role as a synthesizer rather than a replacement for experts
- •Need for consistent, aligned recommendations across domains
- •Focus on actionable clarity vs. more data collection
- 6:11 – 9:57
Building the ‘WellCoach’ GPT: from experiments to a daily-use coaching system
Lucas describes how he began experimenting as soon as ChatGPT launched and gradually turned it into a daily tool. The emphasis is on turning personal data into specific, high-ROI actions that improve recovery and performance.
- •Started with manual aggregation of personal health/performance data
- •Shift from ad-hoc Q&A to a reusable coaching system (custom GPT)
- •Daily usage pattern: decisions around training, recovery, and nutrition
- •Breakthrough value: better synthesis and decision-making
- •Goal: maintain output while reducing breakdown risk
- 9:57 – 16:31
What data the AI coach ingests: imaging, labs, wearables, journaling, and plans
Lucas walks through the actual files he uploads to the GPT, showing how multimodal data becomes usable without heavy structuring. Claire calls out the novelty of including MRIs/X-rays alongside wearables and text-based plans.
- •Imaging: X-rays and MRIs (pre/post surgery) for structural context
- •Wearables (Whoop): HRV/readiness, strain, sleep stages, cycles (CSV)
- •Behavioral context: journal entries (stress, anxiety, sauna, compression, etc.)
- •Clinical inputs: multiple blood tests across time for trend comparison
- •Nutritionist plan + InBody scan for body composition and fueling strategy
- •Multilingual data handling (English + Portuguese exams)
- 16:31 – 17:50
Setting realistic expectations: bounded ambition beats ‘elite at all costs’ prompting
Claire and Lucas emphasize that strong prompting includes both ambition and restraint—clear outcomes, practical constraints, and avoidance of extreme optimization fantasies. The prompt is designed to keep advice accessible, repeatable, and grounded.
- •Outcome-based prompting: define what “success” looks like day-to-day
- •Avoiding extreme ‘most elite in the world’ framing that drives bad advice
- •Practicality filter: accessible actions over exotic therapies (e.g., hyperbaric/ozone)
- •Focus on feeling good, moving pain-free, and sustaining routines
- •Using the model to reduce cognitive load and decision fatigue
- 17:50 – 21:47
Frameworks the coach follows: nutrition, training load, recovery, and feedback loops
Lucas outlines the four operating frameworks embedded in the GPT: nutrition principles, training/load management, recovery as mandatory, and ongoing data-driven feedback. Together, these act like an integrated team of specialists with one consistent strategy.
- •Nutrition: stick to plan unless data suggests adjustment; stable glucose + low inflammation
- •Training/load: balance strength/endurance/mobility; avoid overload when readiness is poor
- •Recovery: sleep as the main lever; PT/mobility/sauna/cold/massage/mindfulness as non-optional
- •Feedback loops: cross-validate across wearables, labs, diet, journal entries
- •Anti-randomness: don’t recommend unaligned internet advice
- 21:47 – 24:25
Hard boundaries and anti-prompts: guardrails against overtraining and low-ROI hacks
This segment focuses on explicit ‘don’t do’ instructions that keep the AI from pushing unsafe intensity or recommending questionable interventions. The guardrails are designed to counter Lucas’s tendency to overtrain and to prioritize proven interventions.
- •No hard training when Whoop shows red/yellow recovery indicators
- •Avoid novelty supplements or unproven tactics; prioritize measurable ROI
- •Act conservatively on red flags (soreness, low HRV, sleep decline, illness signals)
- •Coach must pull Lucas back when ambition exceeds readiness
- •Guardrails reduce hallucination risk and keep advice consistent
- 24:25 – 29:31
Real-life nutrition example: planning around an omakase dinner (and other surprises)
Lucas demonstrates a practical scenario: adjusting the day’s meals to accommodate a higher-carb/alcohol dinner without feeling terrible afterward. The value is rapid, personalized planning and immediate feedback through photos and quick check-ins.
- •Event-based planning: protein-forward meals earlier to offset rice/sake later
- •On-demand adjustments when plans change (party, late night, alcohol)
- •Photo-based meal feedback and small tweaks (e.g., inflammation considerations)
- •Generalizable usefulness: ‘don’t feel crummy tomorrow’ is a universal goal
- •Coach-in-pocket effect reduces friction and increases adherence
- 29:31 – 37:25
Injury management workflow: validating medical advice and building a recovery plan
Lucas shows how he used the coach during an elbow injury by feeding it the doctor’s diagnosis, PT protocol, pain descriptions, and media (photos/videos). The AI helps translate expert guidance into an understandable timeline and checkpoint-based plan, easing anxiety while staying aligned with clinicians.
- •Inputs: diagnosis, PT prescription, daily pain reports, photos/videos of symptoms/range
- •Use case: confirm/validate expert opinions rather than replace them
- •Timeline planning toward a specific competition date with decision checkpoints
- •Reformatting expert advice into digestible, day-by-day guidance
- •Reduced uncertainty: expectations management and anxiety reduction during rehab
- 37:25 – 43:27
Future vision: ambient health data, patient+doctor AIs, and healthcare synthesis at scale
Lucas predicts that everyone will have an AI health coach within years, with passive data capture reducing manual logging. He imagines patient AIs and doctor AIs exchanging context, enabling more informed visits and better between-visit decisions, while top institutions productize guidance via trained models.
- •AI coach as an augmentation layer: better daily decisions and better doctor visits
- •Interoperability and data silos as the key problem AI can help solve
- •Seamless capture: sensors, smart fabrics, potentially advanced biometrics and ambient monitoring
- •Patient AI ↔ doctor AI communication to increase clarity and reduce visit friction
- •Institutional knowledge (e.g., major clinics) potentially offered as specialized AI models
- •Reframing: not gimmicks—precision tools for consistency and population-level impact
- 43:27 – 48:48
Other AI workflows: synthetic clients, synthetic co-founders, and responsible data use
Lucas briefly tours additional work use cases: a “synthetic client” trained on public materials to accelerate consulting work, and an “AI co-founder” for brainstorming in a distributed company. They stress that proprietary client data isn’t used, positioning these as productivity and alignment tools.
- •Synthetic client: approximate stakeholder thinking using vetted, public information
- •Goal: get to 80–90% alignment before consuming busy expert/client time
- •AI co-founder: brainstorming partner for thorny problems when humans aren’t available
- •Distributed work makes on-demand thinking partners more valuable
- •Clear disclaimer: no proprietary client data used—public sources only
- 48:48
Reliability, model evolution, and closing takeaways (AI as a consistency engine)
In the wrap-up, Lucas notes that model quality is improving and that guardrails help manage errors. Claire closes by reinforcing broader applicability (caregiving, kids’ sports, everyday health), and Lucas shares where to find his company while the show signs off.
- •Model evolution: fewer hallucinations over time; errors treated as tech maturity issues
- •Guardrails + bounded instructions improve reliability and safety
- •Broader applications: caregiving, youth athletics, general wellness routines
- •Offer to share the prompt for others to replicate the workflow
- •Final CTAs: where to find Lucas’s company and show subscription/review reminders
Core GPT instructions: role, objectives, and decision lens (protect joints + extend peak)
Lucas breaks down the instruction set that defines how the GPT should think: a performance strategist for a busy operator, not a pro athlete. The coach is trained to prioritize joint safety, pain-free movement, and sustainable performance with minimal fluff.
- •Role framing: “performance strategist and health optimization coach”
- •Primary objectives: safeguard joints, amplify output, extend peak
- •Context constraints: balancing tennis/lifting with company leadership demands
- •Analysis requirements: interrogate recommendations using uploaded data
- •Risk detection: flag red/yellow zones like overtraining, under-fueling, inflammation
- •Preference for high-ROI, evidence-backed actions
Accessibility: making expert-like feedback loops available to more people
Claire highlights that on-demand human coaching is expensive and scarce, while AI can deliver short-cycle guidance and reinforcement. The discussion reframes this as habit support and decision support, not just elite performance optimization.
- •Cost and availability barriers to nutritionists/coaches/PT guidance
- •AI provides quick ‘am I on track?’ feedback loops that reinforce habits
- •Works even with constrained options (e.g., ordering at Chipotle)
- •Behavior change via nudges and contextualization, not lectures
- •Applicable beyond athletes: basics like sleep, movement, less processed food
