Thermal Drift: A Maker's Diary

A silly idea, a patient AI, and fun solo product adventure.

There's a game everyone has played. You hide something. Someone searches. You give one of two signals: warmer or colder. No map, no arrows, no instructions. Just feedback and movement. It's immediately understood, instantly engaging, and — I kept thinking — genuinely more interesting as a way to explore a place than anything I'd experienced from a navigation app.

That thought became Thermal Drift. An iOS walking navigation app that replaces turn-by-turn directions with temperature-based feedback: haptics, audio, and color that intensify as you approach your destination and cool as you drift away. No route line. No verbal instructions. No blue arrow.

I'll be upfront: this project is probably going to serve me creatively more than it serves any actual users. And I'm completely fine with that.

A Silly Idea, On Its Feet Quickly

Thermal Drift started as an experiment — a quick proof of concept to see whether a genuinely abstract navigation mechanic could feel intuitive without being explained. The goal wasn't to ship a polished product. It was to find out if the idea worked at all. What made that possible was AI. I'm not a professional developer. My coding ability is real, but there's a ceiling to what I could build and how fast I could iterate without help. With Claude as a technical partner, that ceiling lifted. I could describe what I wanted the temperature algorithm to do, get working code, run it on a real walk, feel whether it was right, and refine — all in an afternoon. Iteration has never been this cheap.

The pattern that emerged: get to working software as fast as possible, then iterate outward. Core mechanics first. Then stability. Then design refinement. Then performance. AI let me compress the first step dramatically, which meant I got to the interesting part — the feedback loop of real-world testing — much sooner than I would have otherwise.

It's never been more fun to be a maker.

Taking Playful Self Outside

Thermal Drift connects directly to earlier work. Playful Self was a gallery project exploring abstract, non-verbal feedback as a way of engaging people with their own presence in a space — using sensation and surprise rather than instruction. The work asked: what happens when you remove the explanatory scaffolding and just invite someone to move and feel?

Thermal Drift is the same question, taken outside the gallery and onto the street.

What I find interesting is that the "hot and cold" mechanic is universally understood without any instruction at all. Everyone has played the game. The transfer of that knowledge to a navigation context is immediate — no onboarding required. You walk, you feel the phone pulse, you adjust. The game teaches itself.

This is play without gamification. There are no points, no leaderboards, no rewards for speed. The feedback is purely informational, but because it's abstract and physical — you feel it rather than read it — it invites exploration rather than compliance. You want to move around to see what happens to the signal. That curiosity is the product.

Inverting the Feedback Hierarchy

Standard mobile design follows a clear hierarchy: visual first, audio second, haptic third. Haptics are an afterthought — a confirmation tap when you press a button, a buzz when a notification arrives. They almost never lead.

Thermal Drift inverts that. Haptic is the primary channel. Visual and audio are reinforcement.

This was a creative constraint as much as a design decision, and it forced me to engage with the hardware at a level I wouldn't have if I'd taken the more conventional route. CoreHaptics — Apple's low-level haptic API — lets you compose precise vibration events with controllable intensity, sharpness, and timing. I ended up with six distinct patterns mapped to temperature bands: a slow, sparse frozen pulse at long range; a double-beat cold pulse for the wrong direction; a continuous burning vibration as you arrive. The neutral zone — when you're sideways to the destination — is silence. No haptic at all.

Silence as a signal is underused. When the buzzing stops, you know something without being told. That felt right.

Putting haptic first also meant committing to something that works with the phone in your pocket. Navigation that requires you to hold your phone isn't really ambient — it's just navigation with extra friction. The phone goes in your pocket, you feel the pulses through the fabric, and you walk. That's the experience.

The Decisions That Made It Work

A few technical choices were make-or-break.

Logarithmic temperature scaling. Distance is converted to a temperature value from 0 (frozen) to 1 (burning), but the relationship is logarithmic, not linear. This was AI's recommendation, and it was correct. A linear mapping would compress the final 100 meters — where you need the richest signal — into a tiny slice of the feedback range. Log scaling gives you equal expressiveness per order of magnitude of distance. The last block of a walk feels dramatically more alive than the first kilometer. That's the right distribution. No map during navigation. This is the most argued-over decision with testers. Every new user asks for the map. Almost every experienced user stops asking. The map during navigation undermines the whole premise — if you're checking the map, you're not using your eyes to way-find in the environment. The absence of the map is the product. It forces you to trust the feedback.

Testing did surface something real though: there are moments where a user needs to reset their bearings without being handed a map. The haptics hiccup. They've walked for a while and lost their sense of where the destination sits relative to them. Or they're facing a practical obstacle — a building, a construction site, a block that just doesn't look like somewhere they want to walk — and they need to understand the geometry well enough to route around it and keep making progress.

That's what the abstract animation is for. Rather than a map, it shows a visual field that warps toward the destination relative to your current position and orientation — it knows which way you're facing and where the target sits in relation to you. Directional without being instructional. You can see that you need to go around the block to your left without being told to turn left in 200 feet. It stays abstract enough to preserve the drift, but grounded enough to be useful when the haptic signal alone isn't enough to navigate by.

Mystery destinations. Select a category — Coffee, Food, Nature, Culture, Shopping — and the app sends you somewhere nearby. The original idea was to reveal nothing until arrival. I learned quickly that complete mystery is more daunting than it sounds. Testing it in San Juan on a trip to Puerto Rico — a city I'd never navigated before — made that obvious. Knowing nothing about where you're going is fine in a familiar neighborhood. In an unfamiliar one, it's just anxiety. So the feature evolved: you now get the name, category, star rating, and hours before you set off. Enough to make an informed choice, not enough to remove the surprise of actually arriving. That balance — a little information, not full directions — turns out to be the more honest expression of what Thermal Drift is: an invitation to move through a place and pay attention to it, not a commitment to wander blind.

The Honest Part

The stability work has been humbling. GPS on a moving pedestrian is messier than it looks from the outside. Raw distance readings fluctuate 5–15 meters between samples — noisy enough to turn your haptic feedback into jitter if you're not careful. I'm smoothing readings with a five-point rolling average and handling GPS failures with exponential backoff. Mid-route crashes I couldn't explain for weeks turned out to be unhandled edge cases in the location pipeline. Fixing them was unglamorous and necessary.

This is the part AI is less helpful with — not because it can't reason about the code, but because diagnosing intermittent bugs in production requires runs, and real walks, and patience. The technical partner is still useful for structuring the recovery logic. But you have to go outside and test it yourself.

The visual design came together in a way I didn't expect. I had bones — a color direction, a rough sense of the aesthetic — and Claude helped me develop a full design system on top of them in hours. Not days. Refinements that would have taken me a week of trial and error happened in an afternoon. More useful than the speed was the kind of work it unlocked: I was able to build tunable, interactive tools for iterating on abstract animation in the browser, adjusting parameters in real time, before writing a line of Swift. The central navigation visualization — a warped, pulsing field that responds to bearing and distance and syncs with the haptic cadence — exists as a working prototype I can dial in. Low information, immediately readable, strange in the right way. That kind of rapid, hands-on iteration with abstract visuals is something I wouldn't have had access to before.

What I've Learned

I couldn't have built this without AI's help. That's just true. The combination of my creative direction and AI's technical patience got working software into my hands fast enough that I could actually learn from using it. The feedback loop between idea, code, real-world test, and refinement is tighter than anything I've experienced before.

What I'm learning is that the most interesting place to work is inside the constraint. Committing to haptic-first meant I had to understand CoreHaptics deeply, think carefully about silence, and design six patterns instead of one. Removing the map meant I had to make the other feedback channels genuinely informative. Inviting play without gamification meant resisting the pull toward points and leaderboards and letting the sensation itself be enough.

The app isn't in the App Store yet. But every test walk ends the same way: phone in pocket, head up, noticing things I'd have missed with Maps open. That's the thing I was trying to make.

If that's all it does, it's already worked. Thermal Drift is currently in usability testing. More updates as it moves toward App Store submission.