Playground
Experiments, prototypes, and things I'm tinkering with.
Why I Prototype
I've always believed in prototypes. The fastest way to answer a product question is the smallest build that makes it real — something you can feel, test, and pull signal from. And with AI in the toolkit, that build has never been faster or cheaper.
But I don't prototype for prototyping's sake. Every build starts with a purpose — a question to answer, a hypothesis to test, a technology to size up, a friction point to understand. Choosing what to make always carries the cost of what you don't make, so each prototype stays deliberately small, gets in front of real people, and earns its next decision: push forward, change course, or stop.
That's why every prototype below shows its question, what I found, and what happened next.
Passion Project
A standalone AI training companion for runners chasing a sub-three marathon — a full product with accounts, a Supabase backend, and Claude doing the coaching. Decades in the making.
The Question
Could AI finally crack what heuristic training plans never could — real coaching from messy, incomplete runner data — and how far can I take a full AI-native build solo?
What I Found
Separate the brains from the machinery: Claude holds the conversation and makes the coaching decisions; a deterministic plan engine rebuilds the entire 56-week plan in milliseconds via tool use.
What's Next
Sync real running data (Strava, Garmin, Apple Health) and split the single model into specialists — chat, context, coach, reasoner.
Join the chase →
2:59:59
Rapid Prototype
A one-handed basketball stat tracker and a post-game report coaches open minutes after the buzzer — born from a season of pen-and-paper versions at my son's JV games.
The Question
Could I track the stats coaches actually want — turnovers, rebounds, possessions — without taking my eyes off the game, and get a useful report into their hands before they leave the gym?
What I Found
The possession-by-possession view is the unlock — runs, momentum, and control jump out in ways a scorebook never shows. Four games in, the coaches keep asking for it.
What's Next
A few more games of testing. If it sticks, automate the manual workflow — now a known gap I can design to instead of a guess.
Live from the bench →
Hoop Rhythm Tracker
Concept Test
A chat-first onboarding concept for a baby registry, built in a day with Lovable and the AI fully mocked — smoke and mirrors on purpose, to test the concept before the tech.
The Question
Can a conversation replace the one-size-fits-all onboarding wizard — meeting people where they are instead of forcing every edge case down a happy path?
What I Found
Parents, non-parents, people with no kids at all — everyone flowed through. Some went deep, some bailed early, both fine. And testing exposed a path I'd missed: starting a registry for someone else.
What's Next
Clear signal — it earned a works-like feasibility pass, where response speed and tuned suggested answers are the make-or-break.
Say hi to Bee →
AI mocked · On purpose
Technology Exploration
Near-infrared sensors are shrinking toward consumer devices. Two quick Lovable concepts — a skin analyzer and a food analyzer — went hunting for the use case before the tech was even ready.
The Question
Where would a consumer actually want an NIR sensor in their routine — on their skin or on their plate — and is either worth pursuing?
What I Found
Skin won, decisively. Testers tied collagen, dryness, and sun damage straight to product decisions and wanted it at home, not in a store. The food analyzer never found a place in anyone's routine.
What's Next
The sensor hardware isn't ready for a works-like pass — but when it is, the direction is already picked: skin care, at home, connected to the products you buy.
Scan, then decide →
Skin age · 36.2
Internal Tool
One button and one multicolor LED had to communicate power, pairing, connection, errors, and resets. A browser simulator — built with the firmware engineer, carried around on an iPad — became the spec.
The Question
Could we tune every LED color, blink, and press behavior — and catch what confuses people — before the firmware was written or the hardware existed?
What I Found
Walking an iPad around beat iterating in firmware: colors, timing, and messaging tuned live, and testing surfaced cases nobody specced — like an extra-long press for a full hardware reset.
What's Next
The tuned spec went straight to firmware, built once with confidence — and the simulator-first pattern is now the template for the next weakest link.
Enter the simulator →
One button · One LED
Field Test
Pilots couldn't wear the hydration pack — seat fit and regulations ruled it out — so drink reminders had to reach them through their aviation headsets. A bare-bones tool tuned the tones and voices.
The Question
What does a drink reminder sound like in a pilot's headset — clear enough to land, without stepping on the signals that matter for flying safely?
What I Found
Nobody, pilots included, could judge a sound until it played in their actual headsets. Testing climbed from office to hangar to in-flight, swapping tones and voices live against real cockpit audio.
What's Next
The tuned tones and voice went into the app's reminder system — one blocking question answered fast, project on track.
Tune the reminder →
Office → hangar → air
Concept Prototype
A concept for how OpenAI could bring families closer through everyday AI — a landing page, the feature surfaces, and a working Lovable prototype.
The Question
Could ChatGPT feel like it belongs to a whole family — shared memory, shared moments — instead of one person at a time?
What I Found
Framing AI around the family unit unlocked surfaces an individual assistant never suggests — shared memory, carpools, celebrations.
What's Next
Published as a build-in-public walkthrough of how I prototype, from first prompt to working demo.
ChatGPT
Families
A celebration, every day
Product Exploration
An interactive look at why we overeat at meals — calories outrun the feeling of fullness, and the gap between them is the product opportunity.
The Question
Why do we overshoot at meals, and where in the timeline could a nudge actually help someone stop on time?
What I Found
Making the 20-minute lag visible reframes overeating from a willpower failure to a timing problem — the nudge has to land mid-meal, not after.
What's Next
Explore what real-world signal could trigger the nudge — meal timers, eating pace, or a wearable.
Play the meal →
Motion Graphics
The site's logo mark as a motion lab — five experiments in AI-driven animation, section dividers, and brand assets.
The Question
Could I art-direct motion design with prompts alone — no keyframes, no After Effects?
What I Found
Yes — the crackle animation shipped site-wide as a reusable component, and the LinkedIn assets are in real use today.
What's Next
More animated article covers — each new cover type extends the same pattern.
Live — Electric Crackle
Living Prototype
The site you're on right now. After years of Bluehost and Squarespace templates, rebuilt as a hand-built Next.js product — designed, written, and shipped with AI.
The Question
How far past template builders could I take a personal site with AI — and could the workflow keep it genuinely current?
What I Found
The workflow evolved with the tools: Claude Code inside Cursor at first, now a Cowork loop — talk it through, build, review locally, ship. Updating the site went from chore to fun.
What's Next
It's never done — every new prototype, article, and experiment lands here first.
www.steveandthedogs.com
Steve Black
Building What Matters
Clarity. Creativity. Impact.
The site, in the site
From the Archive
Just after Google Glass launched, Adidas and Google went hunting for use cases. The first miCoach-on-Glass prototype wasn't an app — a slideshow of static screens, worn on runs and rides.
The Question
Could real-time coaching live in the corner of your eye — and what could a runner actually read and use mid-workout?
What I Found
Glanceable data thrills, then distracts. Runners needed to pull info with a tap, not have it pushed — except pace-zone moments, which earned the interruption by behaving like a real coach.
What's Next
Those findings shaped the miCoach Glass app before a line of code was written — screens that sleep, wake on workout landmarks, and say only what a coach would say.
Run first, design after →
No code · Just JPEGs