I'm standing at the driving range, AirPods in, phone propped against my bag. I settle into my stance, take a swing. Without reaching for anything, I say: "Coach, 7 iron, pushed that one right." A beat later, a voice comes through my earbuds — calm, specific, familiar with my tendencies: "You've been leaking right with your mid-irons all session. Last week we worked on your alignment at address. Try closing your stance a half-inch and focus on keeping your trail elbow tucked through impact." I adjust. I swing again. No phone. No screen. Just a coach in my ear.
That moment is the entire thesis of OnPlane, the AI golf coaching app I've been building. Not a swing analysis tool. Not a data dashboard. A coach — one that listens, watches, remembers, and knows when to talk and when to let you work.
The Problem with Golf Coaching
A lesson with a good golf pro costs $100 to $200 an hour. Most recreational golfers take a handful of lessons a year if they're serious. But lessons aren't where you get better. You get better at the range, grinding through reps, trying to internalize whatever your pro told you last Tuesday. The problem is that during 95% of your actual practice time, you're alone. No feedback. No correction. Just you and whatever swing thought you can remember.
The market has tried to fill this gap with technology — launch monitors, swing trackers, video analysis apps. They all share the same fundamental flaw: they give you data, not coaching. They'll tell you your club path was 3 degrees out-to-in. They won't tell you why, or what to do about it, or that you've been fighting this same pattern for three weeks and here's a specific drill that might break it. Data without judgment is just a spreadsheet. I didn't want a spreadsheet. I wanted a coach.
Voice-First, Everything Else Second
The core design decision behind OnPlane is that your phone stays in your pocket — or at most, propped up nearby. Everything happens through your earbuds. You talk to your coach; your coach talks back. No tapping, no scrolling, no breaking your rhythm to stare at a screen.
This sounds simple, but it changes everything about how you design an AI product. Most AI apps are built around a chat window — type a prompt, read a response. That interaction model falls apart the moment your hands are occupied and your attention is somewhere else. A golfer mid-practice isn't going to pull out their phone, type a question, read an answer, and put it back. The interface has to disappear. Voice is the only modality that lets you stay in the flow of what you're doing while getting real-time coaching.
Building a voice-first AI product is a fundamentally different design problem than building a screen-first one. Latency tolerance is different — a 3-second delay in a chat app is fine; a 3-second delay in a voice conversation feels broken. Context management is different — you can't scroll up to re-read something, so the AI has to be concise and reference prior context naturally. The whole UX philosophy shifts from "show the user information" to "say the right thing at the right time."
Seeing What You Can't
Voice is the primary interface, but OnPlane also sees. When your phone camera is pointed at you, the app uses computer vision to analyze your swing mechanics in real time. It tracks your body through the swing, identifies key positions, and extracts the biomechanical metrics that actually matter for coaching — the same things a human pro watches for when they're standing behind you on the range.
The difference is that the AI can track these patterns across hundreds of swings without losing focus, and it can correlate what it sees with what you report. When you say "pushed it right" after a swing, and the vision system detected something mechanical that explains it, the coach can connect those dots for you — in plain language, not data tables.
Getting the vision system to work reliably at a driving range — outdoors, varying lighting, different angles, people walking behind you — was one of the harder engineering challenges. Lab demos are easy. Real ranges at 5pm with the sun in your face are a different story. That gap between controlled demo and messy reality is where most AI products quietly die. We made it work.
Knowing When to Shut Up
This is the part most people building AI tools get wrong. They assume more feedback is better. It isn't. If a coach commented on every single swing, you'd fire them. Good coaching means reading the room — knowing when someone is in a groove and shouldn't be interrupted, and knowing when a pattern has emerged that's worth addressing.
OnPlane is deliberate about when it speaks and when it stays quiet. If you're hitting it well, it lets you work. If you ask a direct question, it always responds immediately. But the unprompted coaching — the "hey, I've noticed something" moments — those are metered and deliberate. The system considers your recent trend, how long you've been working on a particular focus, and whether the observation is novel or something it's already told you.
Getting this right was harder than any individual technical component. "When should an AI coach speak?" is a judgment call, and it required a lot of real practice sessions to tune. Too frequent and it's annoying. Too sparse and you wonder if it's working. The sweet spot feels like a coach who's watching closely but respects your process.
A Coach That Remembers
A single coaching session is useful. A coach who remembers every session is transformative. OnPlane maintains a persistent understanding of your game across every practice session. When you show up at the range on Saturday, the AI knows what you worked on Wednesday. It knows you've been fighting a slice with your driver for two weeks. It knows that the alignment drill helped your irons but you haven't tried it with your woods yet.
This is where AI coaching has a structural advantage over human coaching. A human pro sees dozens of students a week and relies on notes (if they take them) to recall your specific patterns. OnPlane has perfect recall. Every shot you've described, every pattern it's observed, every coaching point it's delivered — all of it informs the next session. The advice gets more specific and more useful the longer you use it.
The result is a coaching relationship that compounds. Session one, it's getting to know you. By session ten, it knows your game. By session fifty, it's catching patterns you don't see yourself.
What I Learned Building This
OnPlane taught me something I now apply to every AI project: the hardest problems aren't the models — they're the design decisions around the models. Which model to use, how to prompt it, how to process the data — those are solvable engineering problems. But when should the AI speak? How much context should it carry? How do you make it feel like a coach and not a chatbot? Those are product problems, and they're the ones that determine whether anyone actually uses the thing you built.
OnPlane is currently in beta and launching in Spring 2026. If you're interested in trying it, head to onplane.io.
At BCK Systems, this is what we do. We build AI systems that work in the real world, not in pitch decks. OnPlane isn't a concept or a prototype — it's a production application that runs on real phones at real driving ranges for real golfers. The gap between "AI can theoretically do this" and "here's an app that does it while you swing" is enormous, and it's the only gap that matters. We build across that gap.