Re: chatbots were tried but it’s a wrong model for a deterministic system. It’s the best one for a probabilistic one.

Re: chatGPT auto-router - 1) I suspect explicit model selection matters less in agentic contexts (this thing is coordinating a bunch of other agents for me) vs QA context where you want to make sure you’re getting the best answer. 2) I suspect auto-routing matters less now that models are way better

Comments URL: https://news.ycombinator.com/item?id=49946526

Points: 36

# Comments: 47

This thing in the implementation plan? Change it to that. What did you mean here? See this button? Here is a screenshot. I circled it because it is missing a hover state. This sentence in the third paragraph? Rephrase it.

If you are an opinionated builder, this gets clunky quickly. You either dictate paragraphs of messy feedback or resolve every detail one by one. AI is good at parsing mess, but garbage in, garbage out still applies.

I could get products to 90% quickly. The last 10% of feedback and polish was the bottleneck. So I built Remarc.

Remarc lets you select text in any Mac app, capture and annotate a screenshot, or comment on a web element. Each comment keeps its original context, status, and session. A connected agent can read the session, work through it, update statuses, and leave resolution summaries over MCP.

Think of it like leaving comments on a google doc vs sending a long rambling feedback note via Slack - the more granular you get with your feedback, the more it benefits from structure & better context.

Remarc is free and open source, with no account, subscription, or telemetry.

https://remarc.app/ https://github.com/metedata/Remarc

Comments URL: https://news.ycombinator.com/item?id=49615660

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=49323982

Points: 2

# Comments: 0

There are plenty of tools for collaborating with humans, but surprisingly few for collaborating with AI. Chat was not cutting it for the feedback I wanted to give:

This thing in the implementation plan? Change it to that. What did you mean here? See this button? Here is a screenshot. I circled it because it is missing a hover state. This sentence in the third paragraph? Rephrase it.

If you are an opinionated builder, this gets clunky quickly. You either dictate paragraphs of messy feedback or resolve every detail one by one. AI is good at parsing mess, but garbage in, garbage out still applies.

I could get products to 90% quickly. The last 10% of feedback and polish was the bottleneck. So I built Remarc.

Remarc lets you select text in any Mac app, capture and annotate a screenshot, or comment on a web element. Each comment keeps its original context, status, and session. A connected agent can read the session, work through it, update statuses, and leave resolution summaries over MCP.

Think of it like leaving comments on a google doc vs sending a long rambling feedback note via Slack - the more granular you get with your feedback, the more it benefits from structure & better context.

$0. Remarc is free and open source, with no account, subscription, or telemetry. It requires macOS 14 or later.

Download: https://remarc.app Source: https://github.com/metedata/Remarc

Comments URL: https://news.ycombinator.com/item?id=49323749

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=49314427

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=49303220

Points: 4

# Comments: 0

There are a plethora of tools for collaborating with humans, but surprisingly little for collaborating with AI. Chat just wasn’t cutting it for the kind of contextual feedback I wanted to give my agents.

This thing in the implementation plan? Change it to that. What did you mean here? See this button? Here’s a screenshot, I circled it — it’s missing a hover state. This sentence in the 3rd paragraph? Rephrase it.

If you’re an opinionated builder who wants things in a very specific way, this quickly becomes a nightmare. You either end up dictating paragraphs of messy feedback or resolving everything one by one. AI is great at parsing mess, but garbage-in-garbage-out still applies.

And that’s where I found myself. I could easily get products to 90%, but that last 10% of polish that makes them actually stand out was unnecessarily clunky.

Building with AI has a last-mile problem. So I decided to build something for myself.

Remarc lets you comment on anything on your screen - select any text, any web element, take a screenshot, and leave a comment. Each comment has a status and goes into a session that your agent can start / read / update. Your feedback is now structured and contextual. Comes with webhooks and lots of other bells and whistles so you can dial it in for yourself if this is the way you like to work. Free and open source.

Comments URL: https://news.ycombinator.com/item?id=49300200

Points: 3

# Comments: 0

Social media may make building with AI seem like this: ask for the thing, blindly approve the plan, get the thing. Post “omg claude just one-shotted blah”. Profit.

But if you’ve ever worked on a real product, you know nothing good is ever one-shotted. Every PRD, design, and build goes through dozens of feedback loops and iterations. That’s how you get from something that works to something actually good.

There are a plethora of tools for collaborating with humans, but surprisingly little for collaborating with AI. Chat just wasn’t cutting it for the kind of contextual feedback I wanted to give my agents.

This thing in the implementation plan? Change it to that. What did you mean here? See this button? Here’s a screenshot, I circled it — it’s missing a hover state. This sentence in the 3rd paragraph? Rephrase it.

If you’re an opinionated builder who wants things in a very specific way, this quickly becomes a nightmare. You either end up dictating paragraphs of messy feedback or resolving everything one by one. AI is great at parsing mess, but garbage-in-garbage-out still applies.

And that’s where I found myself. I could easily get products to 90%, but that last 10% of polish that makes them actually stand out was unnecessarily clunky.

Building with AI has a last-mile problem. So I decided to build something for myself.

Remarc lets you comment on anything on your screen - select any text, any web element, take a screenshot, and leave a comment. Each comment has a status and goes into a session that your agent can start / read / update. Your feedback is now structured and contextual. Comes with webhooks and lots of other bells and whistles so you can dial it in for yourself if this is the way you like to work.

Comments URL: https://news.ycombinator.com/item?id=49287672

Points: 3

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=49287541

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=49259972

Points: 2

# Comments: 0

Today I released Remarc - the missing feedback layer between you and your coding agent. Point at anything on your Mac - text, screenshots, web elements, or voice - and your agent reads and resolves your comments over MCP. Think of it like Google doc commenting for anything on your screen.

It's also completely free, local, and open-source.

As a detail-obsessed designer who now builds daily with AI, feedback is a core part of my creative process. AI made it incredibly easy to get from zero to 90%, but it’s the last 10% that sets your product apart from the pack. The unglamorous reality is this 10% is rigorous QA, feedback, and iteration.

Remarc is made for this last 10%. It removes friction to providing structured, contextual, and detailed feedback to your AI agents. My hope is that it helps others craft better products.

Comments URL: https://news.ycombinator.com/item?id=49259711

Points: 3

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=49223115

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=49215695

Points: 4

# Comments: 0