Gurpreet Kalra

Gurpreet KalraSep 27, 2026

Would you sit through a stadium walk-in before a tennis match, or skip it every time? That's the next thing I'm building for Match Point, my solo tennis game on Lovable. Some context first. Pick Wimbledon and you're on grass now. Pick Roland-Garros and you're on clay. I went back in with one question: what would make someone keep playing this, not just try it once? So all four Grand Slams are in, and the court changes with your pick. The US Open and Australian Open put you on hard court. Pick day or night. The night session turns the floodlights on. Difficulty, match format (best of one, three or five) and tiebreaks are all adjustable too. None of that changes the physics underneath. It changes whether you're playing a specific match or a generic rally simulator. Two things are still in motion. The players are getting a pass right now: more athletic shading, more expressive movement. I'm rebuilding that as I write this. Then the walk-in. The referee announces both players by name. A coin toss decides who serves or picks a side. That little ceremony is the moment a match actually starts. Small details, but they decide whether this feels like a tennis game or a tennis simulator. So if you play sports games: is the walk-in and coin toss worth the dev time, or is it the first thing you'd skip? Try it here: https://lnkd.in/gigzdjAm P.S. The walk is a little wonky. Working on that.

Gurpreet KalraSep 27, 2026

Four models. One identical scrambled cube. Same starting position, same instructions. Two solved it. Two didn't, for completely different reasons. Claude Opus 5.5: solved it clean 19 moves · 54/54 stickers 159.5s · $0.31 GPT-6 Sol: solved it, the long way 93 moves · 54/54 stickers 83s · $0.11 Claude Sonnet 5: never finished Ran out of output tokens before producing a final move sequence. 0 moves · 26/54 stickers (the partial state it started from) $0.32 Jev: fast, cheap, wrong Answers one move at a time from the live cube state. No full plan up front. 30 moves · 21/54 stickers 14.7s · $0.0011 Getting the wrong answer cost about 280x less than getting the right one from Opus. That's the number that stuck with me. Nothing here ranks the models. It shows four different modes on the exact same problem: One reasoned correctly and efficiently. One reasoned correctly but took the long way. One ran out of room to finish its own thought. One moved fast and cheap without getting there. Cost and correctness aren't the same axis. Neither are speed and correctness. This is what that looks like on one shared, verifiable task instead of a leaderboard score. If you had to ship one of these, which failure worries you more: Sonnet stalling mid-thought and spending $0.32 for nothing, or Jev giving you a cheap, fast answer that's wrong? Building in Lovable. Will keep experimenting. P.S. The video is trimmed in a few spots. Sonnet spent a long stretch thinking and re-strategizing, so some of that reasoning time is cut for length.

Gurpreet KalraSep 25, 2026

I've always thought of myself as a fast typer. Ask me a few days ago, and I'd have said typing was the most precise way to get what's in my head into actual software. I was wrong. Been building Logoverse, a logo generator that came out of a specific problem I kept running into myself, still very much in progress, getting close to launch. Today I tried building it using voice mode in Lovable's newly released 'chat' feature instead of typing. Typing has a cost I never noticed until I stopped doing it. You filter your thought before you hit enter. You compress something you understand fully into a short, rigid instruction. A few hours in, your hands get tired and your ambition quietly shrinks to match. Talking through it out loud skipped all of that. Explaining a layout choice, catching an edge case mid-sentence, riffing on an interaction, none of it needed to be compressed first. It just came out as a real conversation with the build, not a command typed at it. This is the first time "vibecoding" actually felt true to its name for me. Not because it's hands-off, it isn't. Because my actual thinking reached the software without getting stuck behind my fingers first. Logoverse is coming along. But the bigger thing I'm sitting with: I used to think the keyboard was just how you build. Now it feels like the compromise it always was. Tried voice-first building yet? Curious what it changed for you, if anything.