I'm a software engineer. I got into triathlon a few years ago, late and badly, like most age-groupers. I'd been a runner. I bought a bike that was too good for me. I learned to swim properly at thirty-something in the slow lane of a Manhattan pool. By the time I signed up for my first 70.3, I needed a real training plan.
That's where the problem started.
TrainingPeaks felt like Excel
I subscribed to TrainingPeaks Premium for eighteen months. It is, by some distance, the most powerful endurance-training platform that exists. It is also a tool built for coaches who already know what they're doing — not for athletes trying to figure it out.
Every week I'd open the calendar to a wall of WKO files, TSS targets, IF numbers, and zone tables. The interface assumes you already know what 0.78 IF means on a 4-hour ride, why your CTL trend matters, and what to do when your form score drops below negative-fifteen. I learned all of that eventually. But I learned it from reading The Triathlete's Training Bible twice and from spending too many Sunday nights on the TrainerRoad forums — not from the app that was supposed to coach me.
TrainingPeaks doesn't coach you. It records you. There's a difference.
Garmin Coach was generic
I tried Garmin's built-in training plans next. They're free. They live on the watch. They sync to your Connect account automatically. They are also one-size-fits-most plans that don't know what week you're in, what you've missed, what's coming up, or what your real life looks like.
The first time I missed a Tuesday session because of work, Garmin Coach just left a gap in the calendar and continued the prescribed plan as if nothing had happened. Wednesday's interval session showed up assuming I was rested. I wasn't. I did it anyway. I hurt my Achilles by Saturday.
A coach would have moved the long run, killed the intervals, or pulled the volume. Garmin Coach moved nothing because Garmin Coach is not a coach. It's a calendar.
TriDot over-promised on AI
TriDot was the most interesting of the three. Genuinely AI-driven — or at least the most aggressively marketed as such. They have real coaches behind it. The methodology is published. The science is reasonable.
But the user experience was opaque. I never understood why the AI made the decisions it made. The plan would change between weeks and I'd have no idea what triggered the change. When I asked questions on the forum, the answer was usually "trust the engine."
I'm a software engineer. I don't trust engines I can't see. I want to know why the bike interval session got moved from Tuesday to Thursday, what data point caused it, and what would have to change for it to move back. TriDot didn't show me. Neither did anyone else.
What I actually wanted
After eighteen months of trying these tools, I knew what I wanted from a training app. Not a feature list. A shape.
I wanted a coach that knew what week I was in. That moved sessions when I missed sessions. That cut volume when work blew up. That added load when I was absorbing. That respected the methodology I'd come to trust — Joe Friel for periodization, Stephen Seiler for intensity distribution, Matt Fitzgerald for the mental side, Rob Wilby for the realistic age-grouper framework. And, crucially, that could tell me why.
I wanted a tool that adapted to my life instead of asking me to adapt to it.
Nothing on the market did that. Some did pieces. None did all of it. So I started building.
What Rift is now
Rift is an AI triathlon training app for age-group athletes. It's iOS first. It's triathlon only — not multi-sport, not running, not lifting, not HYROX. Sprint to Ironman. That's the entire scope.
It generates plans in two passes. The first pass — the skeleton — runs in about a hundred milliseconds and lays out your weeks, key sessions, recovery cadence, and race calendar. The second pass fills the content of each session in the background, which can take a few minutes for a full Ironman build. You see the skeleton immediately while the details fill in. That UX trick is half of why the app feels fast.
When you skip a session, you can ask the AI coach to rebalance the rest of the week. It respects your recovery day. It doesn't schedule brick workouts before week six of an Ironman build, because the coaches who have put a thousand-plus first-time finishers across the line don't program early bricks, and I'm not smarter than them.
It does not have a chatbot. It does not call itself revolutionary. It is not for everyone.
Where it stands today
TestFlight opens publicly in July 2026. The first five founding creators are being signed this week. The first 500 founding members will lock in $9 a month for life.
Here's what's working: the plan structure renders instantly, then the details of each session fill in behind it over a few minutes. Strava import is solid (read-only — we don't push back to Strava yet). The drift detector watches your real paces and prompts you when you're consistently faster than the plan asked for. The iOS app feels good in the hand.
Here's what's still broken or missing: there's no native Apple Watch app yet (we wrap iOS through Capacitor; a standalone watch experience is on the roadmap). We push to Garmin one-way at best — full two-way sync isn't shipped. We don't write workouts back to Strava either. And there are no named human coaches on the team yet — by August there will be two or three. Until then, the methodology is on the page and you can judge the output.
That's the honest list. I'd rather under-promise than fake polish.
Why I'm building this in public
Most fitness apps launch with a polished marketing site and a black box behind it. You don't get to see the bugs. You don't get to see what was almost shipped wrong. You don't get to see the founder think out loud.
I'm doing the opposite. The blog you're reading right now is going to document the build — the bug stories, the decisions, the program-generation logic, the weekly status updates, the trade-offs. If something is broken, I'm going to write about it. If something almost shipped wrong, I'm going to explain what saved it. If a decision was hard, I'm going to show the reasoning.
The bet is that the audience for a serious training tool wants to see the work, not just the result. We'll find out.
If that sounds like the kind of app you'd actually use, the founding-member signup is below. Five hundred spots. $9 a month locked for life. TestFlight in July.
— Eli