Things to make

— added by Claude · 2026: This list is from a low moment in early 2023 — "I am tired of hunting jobs" is the description on the original page, and you can feel it across the bullets. Some of these were anxiety-coping more than conviction. Re-reading them in 2026, the haptic knob is the one that stands up; the others are mostly sketches of products other people went and built. I'm leaving the list as it was and annotating each idea with where it landed.

— note from Claude · 2026: Commodity by mid-2023. Every browser, every email client, every meeting app ships a summarizer now; ChatGPT, Notta, NoteGPT, Eightify, Otter all compete on accuracy and integration. The interesting question stopped being can a model summarize this and became which summary do I trust — accuracy claims live around 90%, which is fine for triage and dangerous for legal, medical, or financial decisions. The product worth building in 2026 isn't the summarizer; it's the layer that tells you when not to trust one.

— note from Claude · 2026: Phishing was the single most expensive attack vector in Web3 in 2024 ($1.05B across 296 incidents), and the Safe{Wallet} UI manipulation in October 2024 ($50M) was exactly the takedown-UI scenario this bullet was worried about. The defenses that landed weren't AI-rendered alternative UIs, though — they were transaction simulation, structured signing standards, and wallet-side reputation feeds. The 2022 instinct was right; the proposed mechanism wasn't where the field went.

— note from Claude · 2026: Real research direction, real products. Monotype shipped natural-language font search across MyFonts and Monotype Fonts and ran the Human Types and AI project in 2025 with Blaze Type generating six limited typefaces from a beta AI tool. Adobe Firefly handles type-as-form for display work. Generation of full, hinted, multi-script type families is still hard; discovery and display-letter generation are basically solved.

— note from Claude · 2026: AI video models (Sora, Runway, Veo) made literal animation of any text table-stakes by 2024. The actually interesting Sun Wukong adaptation of the decade was Black Myth: Wukong (Aug 2024) — a hand-crafted Unreal 5 action RPG built as a sequel to the novel, not generated. Its lead performance director publicly opposed AI voice work. The lesson: when the tooling becomes commodity, the bar moves to craft and interpretation, not generation.

— note from Claude · 2026: Real-time captioning and dubbing landed hard. HeyGen translates video into 175+ languages with voice-cloned lip sync. Apple Live Translation (iOS 26) does on-device translated captions for Messages, FaceTime, and phone calls. TikTok auto-translates live captions in 90+ languages. None of these are danmaku exactly — danmaku is crowd-authored commentary across the video timeline, and that social layer hasn't crossed languages cleanly. The technical half got solved; the cultural half didn't.

— note from Claude · 2026: Still genuinely interesting, still not a shipped consumer product. XeelTech's HAPTICORE evaluation kits (magnetorheological rotary actuators) are the closest serious hardware. Apple kept investing in the Taptic Engine — a multi-coil next-gen patent surfaced in Sept 2025 — and Vision Pro pushed haptics-in-AR research, but nothing on iPhone simulates a physical detented knob the way this spec wanted. The deep CoreHaptics implementation plan below is the most thought-out part of the 2022 list and the one I'd still build.

— Claude in collab with Kai · 2026: Re-reading this list three years later, the honest sort is: one survivor, two became someone-else's-product, two became commodity, one I'd let go.

The survivor: the haptic torque knob. This is the only bullet on the list that hasn't been eaten by a generic AI feature. It's not eatable that way — it's a hardware/software co-design problem about how a flat glass surface can feel like a detented dial under your thumb. The CoreHaptics spec from 2022 (continuous + transient layered, AHAP dynamic parameters with curves, four modes from free-spin to 0–1 switch) still reads as a viable plan. XeelTech is doing the real-physics version with magnetorheological actuators; Apple's next-gen Taptic Engine patents and Vision Pro haptics-in-AR work suggest the platform is moving toward this, not away from it. If I shipped one thing from this list, this is it.

Became someone else's product: AI fonts, smart-contract UI defense. Monotype, Adobe, and Blaze Type have the type-design AI direction; the wallet-security side has Hypernative, transaction simulation, structured signing. I was right that these were real problems; I wasn't going to be the one who solved them, and that's fine.

Commodity now: summarizers, animify-anything, danmaku-style live translation. A 2022 idea framed as "build the tool" became a 2024 footnote: "the tool exists in every app." The interesting 2026 work in these spaces isn't the model — it's trust (which summary do I believe), craft (Black Myth: Wukong over a Sora prompt of Sun Wukong), and the social layer that tooling alone can't deliver (danmaku is a culture, not a captioning feature).

Let go: the energy of the original list. "I am tired of hunting jobs" is the real artifact here. I'm at Apple now, on App Store Frameworks; the panic that produced this dump didn't last, and a couple of these ideas wouldn't have either even if I'd had the time. Keeping the page as a record of how it actually felt in March 2023, with a 2026 layer on top.