How I AIHow to build an iPhone app with zero technical skills
CHAPTERS
- 0:00 – 6:16
Daily Hundreds: from one prompt to a working MVP
Bryce explains how she—despite being non-technical—prompted Lovable/Replit to generate a basic MVP for Daily Hundreds. She shares the original motivation: a simple daily habit of doing 100 reps, made less boring by varying exercises and tracking progress.
- •Daily Hundreds concept: a daily prompt to complete and log 100 reps
- •Initial build happened remarkably fast from a minimal prompt
- •Chose Replit as the primary build tool after trying Lovable
- •Personal origin story: pandemic inactivity and wanting lightweight workouts
- 6:16 – 8:14
Replit workflow for non-devs: previews, iteration, and plan mode
Bryce and Claire unpack how Bryce actually works inside Replit day-to-day. The big unlock is learning to use planning/structured steps to avoid the tool making “bananas” changes and to keep iteration controlled.
- •Beginner’s mindset as an advantage: not knowing constraints encourages experimentation
- •Using Replit’s plan/structured approach before making changes
- •Preview panel as rapid feedback for whether changes worked
- •Undo/rollback mentality when the tool veers off track
- 8:14 – 11:43
What she learned (and didn’t): ‘vibe coding’ and infrastructure by trust
Claire probes what technical skills Bryce developed, and Bryce candidly says she didn’t learn much traditional software engineering. The conversation highlights a new reality: non-developers can adopt developer infrastructure (like Railway) via AI-guided steps without fully understanding it.
- •Minimal ‘hard skills’ gained; emphasis on organization, labeling, and copy/paste
- •Shipping progress without deep comprehension of hosting/infrastructure
- •Railway adoption as an example of AI lowering dev-tool complexity
- •Asking for ELI5 guidance and using screenshots to navigate confusion
- 11:43 – 12:56
Why add exercise demos: users need guidance, not just a rep counter
Bryce describes early user feedback: people asked what exercises meant and how to do them. To remove herself as the bottleneck, she decides the app needs demos—and lands on the creative hook of anthropomorphic animals instead of human-only videos.
- •Users asked basic questions: ‘What’s a superman?’ ‘How do I do reverse lunge?’
- •Need to avoid being the “single point of failure” for explanations
- •Creative decision: anthropomorphic animals as fun, scalable demos
- •Commitment to building video coverage for nearly all exercises
- 12:56 – 15:12
Video pipeline overview: Sora → Gemini images → Higgsfield + Kling motion
Bryce walks through the evolution of her production workflow after Sora proved limiting. Her final pipeline uses Gemini to generate the starting animal image and Higgsfield’s motion transfer (Kling model) to combine a still image with Bryce’s real exercise video.
- •Sora produced inconsistent results and limited fine-tuning
- •Gemini used to generate high-quality still images (starting positions matter)
- •Higgsfield used to merge still image + motion video (motion transfer)
- •Kling model worked best inside Higgsfield for this use case
- 15:12 – 20:07
Live prompting in Gemini: precision, anatomy, and “barre teacher” cueing
They create a new animal (a leopard) and Bryce demonstrates how detailed prompting drives better posture and alignment. Claire points out Bryce’s hidden advantage: exercise cueing skills translate directly into precise prompt instructions.
- •Choosing an animal + exercise combo that matches a real starting position
- •Writing hyper-literal prompts (knees over hips, feet in tabletop, elbows out)
- •Iterating when the model ignores constraints (hands/legs/head position)
- •Career skills transfer: fitness cueing becomes prompt-engineering leverage
- 20:07 – 23:27
Iteration tactics: restart prompts, add constraints, and prevent “extra friends”
After the first generated leopard image is flawed, Bryce demonstrates her iteration playbook: rewrite instead of tweaking when stuck, add explicit constraints, and reduce ambiguity. She also shares common failure modes like the model introducing extra characters or unwanted props.
- •When outputs are wrong, rewriting from scratch can outperform minor edits
- •Add constraints: ‘both feet off the ground,’ ‘no other characters’
- •Common model drift: adding random characters, weights, or wrong positioning
- •Using reference photos/screenshots when the model won’t converge
- 23:27 – 25:41
Turning the image into video in Higgsfield: motion control setup and patience
Bryce shows the exact Higgsfield configuration: Motion Control with Kling 3, selecting the exercise video as motion and the generated image as the scene. They discuss practicalities like queue limits, long generation times, and keeping the gym environment consistent.
- •Higgsfield path: Video → Kling 3 → Motion Control
- •Pick motion source (Bryce’s recorded crunch video) + scene source (leopard image)
- •Scene control set to image to avoid showing the living room background
- •Generation constraints: time, queueing, and waiting between runs
- 25:41 – 29:57
Anthropomorphic highlight reel: successes, failures, and the floating genie
While waiting for the new video render, they review past creations—animals and mythological creatures—and where AI goes hilariously wrong. Bryce explains how subtle starting-position mismatches cause major motion-transfer failures, illustrating why the initial image is the true foundation.
- •Showcase of different characters (goblin, fox, flamingo, genie, turtle)
- •Example failure: genie ‘Supermans’ with bizarre floating/limb artifacts
- •Example success: turtle performing multiple movements surprisingly well
- •Key insight: accurate starting images dramatically improve motion results
- 29:57 – 32:16
Leopard crunches result: high-production output from simple inputs
The generated leopard crunch video comes back and works—complete with natural movement and even a plausible mirror reflection. Claire emphasizes how this kind of production value previously required specialized animation resources, validating the power of the workflow.
- •Final video matches the crunch motion convincingly
- •Notable realism: reflections and slight ‘struggle’ wobble
- •Contrast with pre-AI baseline: low-res home videos or expensive animation
- •Workflow recap: image prompt + phone video → polished character demo
- 32:16 – 36:25
From Replit web app to iPhone App Store: Claude planning + Claude Code execution
They shift from content creation to shipping: Bryce explains how she navigated App Store submission without mobile dev experience. Her approach: use Claude for high-level planning, Claude Code for implementation, and then run terminal commands herself to bridge permission gaps.
- •Early advice said she’d need a technical person—later models made solo possible
- •Prompt to Claude: ‘How do I prepare a Replit app for App Store submission?’
- •Workflow roles: Claude as architect/PM, Claude Code as engineer, terminal for final steps
- •Intense sprint: ~25–30 hours in a weekend to reach submission readiness
- 36:25 – 37:58
App review iteration: fixing checkboxes, Sign in with Apple, and account deletion
Bryce describes what Apple rejected on the first submission and how she resolved it quickly by pasting the feedback into Claude. The fixes are practical compliance items: correct parental settings, ensure Sign in with Apple works (especially for iPad compatibility), and add account deletion.
- •Apple feedback copied into Claude for guided debugging
- •Parental/child-safety setting was the wrong checkbox (submission metadata issue)
- •Required: working Sign in with Apple (and actually testing it)
- •Required: in-app account deletion button for compliance
- 37:58 – 46:33
Lightning round: hiring in the AI era, book recs, and prompting discipline
They close with Bryce’s reflections on how roles and hiring expectations shift when “robots can find a working solution faster.” She recommends books about adapting beyond old success patterns, and shares her pragmatic prompting approach: be literal, reset, try new modalities, and keep composure (most of the time).
- •Hiring and being hireable: humility, cross-pollination, and role redefinition
- •Engineers’ interviews should emphasize broader judgment beyond fast solutions
- •Book recommendations: ‘What Got You Here Won’t Get You There,’ ‘How Women Rise,’ ‘A Whole New Mind’
- •Prompting tactics: hyper-literal instructions, restart/refresh, screenshots as guidance
