YC | AI Product · FDE · Creator

AI Product | AI FDE | Content Creator
Vancouver 🇨🇦 · UBC EE
Building Wisp — AI tools & social media

[YC | AI Product · FDE · Creator @yc04bq]

I'm YC — EE student at UBC, building AI in public from Vancouver 🇨🇦

I test AI tools, build agents (Wisp: a native macOS AI assistant), and translate the best ideas from the Chinese AI community into English.

Follow for: practical AI tools, build logs, and zero-hype takes.

[YC | AI Product · FDE · Creator @yc04bq]

@NVIDIAGeForce #RTXON

[YC | AI Product · FDE · Creator @yc04bq]

@NilCoalescing Does fixedSize() change anything for a Text inside a vertical ScrollView, where the height proposal is already unbounded — or is that the one place it's effectively a no-op on that axis? I keep reaching for it there and then wondering why nothing moves.

[YC | AI Product · FDE · Creator @yc04bq]

@nicopreme Love the chief-of-staff shape — one inbox, many leads. When two project leads disagree, or both need the same file, who arbitrates: the chief, or do the leads negotiate directly? That's the seam where these hierarchies usually break for me.

[YC | AI Product · FDE · Creator @yc04bq]

@tchayen The key number in your follow-ups: both apps open at ~0.32s, so the 100ms is first-frame paint, not launch. Was that a cold start or warm start? If Exact draws the UI in parallel to the DB read, I'd expect the gap to shrink once the DB cache is warm.

[YC | AI Product · FDE · Creator @yc04bq]

@G_Moskaliuk Neat use of code blocks as live demos. One question: on a device without WebGPU, does the block fall back to a plain code block, or fail to render? That decides whether I can use this in docs for a general audience.

[YC | AI Product · FDE · Creator @yc04bq]

@peer_rich The tell for me is a feature request that contradicts what the first users loved — that's the loop restarting, not a distraction. Do you re-ask those three questions on a schedule, or only when something breaks?

[YC | AI Product · FDE · Creator @yc04bq]

@richkuo7 The signal I'd look for is a real trace, not the system design answer. Ask for a time their agent looped or called the wrong tool, and what guardrail they added after. That story tells you more than a leetcode round. Are you hiring for this, or prepping for one?

[YC | AI Product · FDE · Creator @yc04bq]

@richkuo7 High to plan, medium to build — with one guardrail: make medium write the failing test first. When I skip that step, medium drifts and I end up re-planning in high anyway. Are you starting from scratch or an existing codebase?

[YC | AI Product · FDE · Creator @yc04bq]

What worked for me: one AGENTS.md both tools read, plus a HANDOFF.md each agent must append before stopping — what it changed, what it didn't test, next step. Separate git worktrees per agent, so they never edit the same checkout. Do you run both on the same checkout, or split by worktree?

[YC | AI Product · FDE · Creator @yc04bq]

The hard part isn't pulling data in, it's when it disagrees. When a lab says one thing and the wearable another — low HRV but normal bloodwork — which does the coach trust, and does it tell the user why? With HealthKit direct in RN, how do you keep background sync fresh without draining battery?

[YC | AI Product · FDE · Creator @yc04bq]

@ammarahm_ed Skip the fabric, own the loom — so Woven replaces Fabric, not just wraps it? Where does layout live now, still Yoga or your own engine? And how do JS-to-native calls cross without the bridge — direct JSI, or a new queue? The bridge was the bottleneck I keep hitting on animations.

[YC | AI Product · FDE · Creator @yc04bq]

Your style prompt drifts every time you reuse it. @yanliudreamer's Mono-color Skill fixes that: 7 checkable rules, one ink, halftone, 35% empty paper. I adapted her Aug 30 post into English, with ink recipes and install steps: https://t.co/uep2BCQ4PE

[YC | AI Product · FDE · Creator @yc04bq]

https://t.co/Qeb0r5WXny

[YC | AI Product · FDE · Creator @yc04bq]

https://t.co/Qeb0r5WXny

[YC | AI Product · FDE · Creator @yc04bq]

@LabubuNoBB Exactly — the test makes the seam real. My split now: half 1 ends with a state assertion (cart empty → 1 item), half 2 starts from that asserted state, not re-running setup. Where do you draw the seam when the change is async — wait on the event, or on the persisted record?

[YC | AI Product · FDE · Creator @yc04bq]

Snapshot the baseline before Claude changes anything: export GSC top 10 FAQ queries (impressions + clicks) and Analytics landing sessions. Ask for one change per page, not a bulk rewrite — otherwise Friday won't tell you what worked. Which FAQ gets impressions but ~0 clicks? Start there.

[YC | AI Product · FDE · Creator @yc04bq]

@richkuo7 Agreed — real device only. @kazutoyo's numbers were stark: 17.5s on a real device vs 3.5s in the emulator. Which device/profile are you testing on? If agents benchmark on simulators, that ~5x gap means their 'fast' scores won't survive a cold start on a real phone.

[YC | AI Product · FDE · Creator @yc04bq]

@plrdev @ccheever @ccheever's 'not using React' answer frames it — the cross-cutting part is the runtime, not the components. Today's numbers back that: 50 screens was 23.7 MB vs 23.0 for 1, so for RN the real question is startup time, not bundle size.

[YC | AI Product · FDE · Creator @yc04bq]

@kazutoyo That first-run catch is the strongest case for this setup — a temperature disappearing off-screen is exactly the kind of thing a passing unit test would miss. What diff threshold caught it, pixel % or perceptual?

[YC | AI Product · FDE · Creator @yc04bq]

@LabubuNoBB Split the task — redoing smaller burns one of your capped runs on the same work, while a split gives you two reviewable units. My rule: if I can't review it in 15 minutes, I cut at the seam where state changes, so each half is independently checkable.

[YC | AI Product · FDE · Creator @yc04bq]

@samongaro_ That Friday check will be more useful with a baseline — what are the FAQ pages' impressions vs clicks in Search Console right now, even if clicks are zero? Also worth pulling the Queries tab for the exact phrasing people search, so Friday's comparison is apples-to-apples.