Show HN: Realwork – Proof of work for the AI era | Hacker News Reader

Show HN: Realwork – Proof of work for the AI era

realwork.app

1 pointAnuranjan_Vikas6 comments

Today, AI can generate anything from text, art, code to even music. The output is often indistinguishable from human work. And the people getting hurt aren't just consumers of slop, it's creators who now have to prove they didn't use AI.

Writers avoid em dashes because they "look AI." Artists get accused before they're appreciated. Students get flagged by inaccurate detectors.

Realwork is a macOS app that records your creative process: keystrokes, pauses, revisions, the messy parts and generates cryptographic proof that your work was shaped over time, not generated in seconds.

Think of it like a digital witness. You get a shareable proof page showing the timeline of how something was made. No surveillance, no productivity tracking, just receipts when you need them.

Tech: Native Swift/SwiftUI, ScreenCaptureKit for recording, Vision Framework OCR, blockchain-style session blocks with SHA256 hashing + Secure Enclave signatures.

I wrote the manifesto with Realwork running: Manifesto: https://www.realwork.app/manifesto Proof: https://www.realwork.app/anuranjanvikas31/launch-blog

Would love feedback on the concept, the UX, or whether this even matters to you. What would make you actually use something like this?

Open articleView on HN

I am struggling to understand the need for this. If AI output is indistinguishable, that could mean -

1. It is as good as human output. For this to happen with the current state of AI, a human still needs to be an expert to guide and review. In this case, I don't think it hurts anyone to use AI.

2. It is worse than human output but the humans consuming it don't know any better. The AI reproduces successful patterns to produce books, paintings, music, films, instead of producing novel ideas/concepts that add to society. This is a deeper problem, and humans in pop culture have been doing it for years without AI. And besides art, bad outputs like vulnerable code or inaccurate text get detected naturally sooner or later.

You're right that the quality question is nuanced. But Realwork isn't really about whether AI output is good or bad—it's about what happens to creators in the meantime.

The problem I'm trying to solve: a student writes an essay, gets flagged by a detector, and now has to prove a negative. An artist posts work, gets accused in the comments, and has no way to show their process. The burden of proof has shifted to creators, and there's no good tool for that. You could argue "good work speaks for itself" and historically that was true. But we're in a transition period where the default assumption is shifting. Even if AI detection eventually improves or people learn to distinguish quality, right now creators are getting caught in the crossfire. Realwork is less about "AI bad" and more about giving creators receipts when they need them. Whether that need persists long-term, I'm genuinely not sure. But the pain is real today.

Curious, do you think the accusation climate is temporary, or does it become a permanent feature of how we evaluate work?

Makes sense. The default assumption is indeed shifting. Today, you are guilty of using AI until proven innocent. Students and some artists might certainly benefit from what you demoed.

This is interesting, I'll play the devil's advocate a bit more.

1. What is stopping me from opening AI outputs on another screen and typing them myself?

2. Have you figured out how to extend this concept to music and photography?

Good questions and Thanks for asking.

1. Honestly? Nothing. You could type out AI output from another screen and Realwork would record it. The friction helps, most people won't bother, but it's not foolproof. The typing patterns might look different (copying vs composing tends to show), but a determined faker could game it. At some point, if someone's putting that much effort into faking authenticity, that's a social trust problem no tool fully solves. Realwork raises the floor, not the ceiling.

2. Digital art (Figma, Photoshop etc) works already, recording captures the visual process, strokes, layers, iterations. Music and photography are trickier. Music production (Ableton, Logic) could work by tracking DAW activity, but still figuring out what "proof" looks like there. Photography is hardest since it's capture, not creation over time, though editing in Lightroom could apply. Some mediums will translate cleaner than others.

ironically, this could maybe sell for ai training.

Lol I've thought about that too. If we ever get there, user controls their data, they'd decide if they want to sell it or not. Worse pivots exist.

Reply on news.ycombinator.com