geooff_

Where I grew up there was no "Science Fair Circuit" like described in this article. Science fairs where a way for young kids (aged 8-10) to test silly hypothesis. There was no feeding into national fairs or anything like that.

I remember one being how fast bean sprouts grew when watered with different liquids (Water, Olive Oil, Wine, Coke, ect). An idea a kid came up with and tested with only minimal help from parents.

To me they should be about exploring independent ideas. I love the Adam Savage quote: "Remember kids, the only difference between screwing around and science is writing it down". To me this is what they should capture.

It's about preserving brand image. Destroying a product is favourable compared to selling it at a discount and making the brand you spent so much marketing appear "cheap".

We need a standard way of identifying agents/bots in the footers of posts. I even find myself falling for this. I use Claude Code to post a comment on a PR on behalf of myself, but there's nothing identifying that it came from an agent instead of myself. My mental model changes completely when interacting with an agent versus a human.

It's refreshing to see a realistic (and familiar) product story laid out. I've gone through this exact process (Thankfully I didn't try to self-host email) trying to launch products.

Congratulations on getting it out the door!

Maybe I'm naive, but I find these re-engineering complex product posts underwhelming. C Compilers exist and realistically Claudes training corpus contains a ton of C Compiler code. The task is already perfectly defined. There exists a benchmark of well-adopted codebases that can be used to prove if this is a working solution. Half the difficulty in making something is proving it works and is complete.

IMO a simpler novel product that humans enjoy is 10x more impressive than rehashing a solved problem, regardless of difficulty.

It doesn't interact with sim. You still need XcodeBuildMCP for that. Hopefully future releases implement this functionality.

XcodeMCP (Native MCP added in 26.3) Implements this with RenderPreview

RenderPreview: Builds and renders a SwiftUI #Preview, returns snapshot

100% I hope they open more of the tooling to MCP, Xcode Instruments with real MCP support would be huge.

This is huge news. Human-in-the-loop development is essential for actual software velocity gains. The current tooling around agent enabled iOS dev leaves a lot to be desired. Every time I work on web-dev tasks I'm jealous of the tooling.

I was refreshing new checking to see I wasn't the only one haha

Having dictation and worktree support built in is nice. Currently there is a whole ecosystem of tools implementing similar functionality for Claude Code. The automations look cool too!

Xcode truly is terrible.

This article has a catchy headline, but there's really no content to it. This is content marketing without content. It seems like every week on Hacker News, there's a dozen of these. All seemingly code reviewers, too. Keep it to LinkedIn.

This is just an ad. It doesn't even offer any interesting data, just a plausible architecture. I had to remind myself I'm not on LinkedIn

One thing to call out, in my experience Coding Agents are bad a SwiftUI. Compared to doing JS frontends or any Python Backend work the difference is obvious.

I just don't think theres enough Swift in the LLMs corpus and the "right way" to do things in Swift has changes a few times in the last few years which I imagine compounds the sparse signal.

For SwiftUI work I'd personally recommend using Opus 4.5 with Axiom. Anytime you're designing something refer to Axiom, Claude needs its skills and agents to steer designs.

It would be nice to see more than 2 opening moves before getting login blocked.

It's only getting worse with the amount of AI Slop being poured into the store.

I'm relatively new to the space, but it feels like more and more of the time of indie devs / bootstrappers needs to get allocated towards marketing.

This is huge. As someone who's put together a somewhat cohesive product but wants to take the next step in "formalizing" components there is a huge missing piece for design.

If I know my backend works I can use claude to add sufficient tests and iterate from there, improving code maturity. If my UX looks good how can I formalize it and do the same?

I can't wait to try this out. My current workflow has been use MCP to screenshot the whole app --> Use a tool like Codia to go from screenshot to components, as a non-designer its miserable though.

Glad to see he's still the same guy

Well written. One interesting pattern in founders I've been thinking about recently is how early you encounter major blockers, and how that shapes your self worth.

Having a strong product and being unable to raise feels like shit, early users may validate you're solving a problem but it may not make sense to investors.

Having an early raise is a very positive signal, I can only imagine you're perceived trajectory coming out of that. Being unable to settle on a problem also sucks in a different way.

Regardless of what camp you're in I feel the take-away is you need to focus on reflecting inwards. Be better than you were yesterday, not better than someone's projections on LinkedIn.

I think the original post was just headline bait. There is such a fast news cycle around AI that many people would take "Thousands of AI agents collaborate to make a web browser" at face value.

I’m a remote dev building an iOS app called Springus. You add a few pics / your closet basics, and it suggests outfits based on what you already own. It’s less “fashion app” and more “tiny morning momentum hack.”

Would love feedback from other WFH folks — a weird amount of my productivity comes down to how I start the day.

This is super cool. Using LLMs to mix together two songs that don't belong mixed together is a great usecase.

Very cool. Is there an option to self-host? This seems like it could be a cool agent skill.

Their published benchmarks leave a lot to be desired. I would be interested in seeing their multi-image performance vs. Nano Banana. I just finished up benchmarking Image Editing models and while Nano Banana is the clear winner for one-shot editing its not great at few-shot.

Seems like theres no API access. Has anyone got the weights? I'm not sure what to fill out for `affiliation`

I'm looking for AI features in a few places, one example is a git client that can draft commit Summary and Descriptions. This should be a very simple task and could use a simple on-device model. It feels like AI features are a firehose though. You either get none and a very 2018 product experience or a complete rework of everything "designed for agents".

After failing to create a hub (too large in my cade) the state of hub selection isn't reset (you need to cancel manually before retrying)

Extracted features from a vision model. I haven't experimented with CLIP yet but would like to as I think adding clothing search would be interesting

The app can handle everyday shots, as you'd expect though, poor inputs produce poor outputs. Theres really two components to this question though:

1. Can the app differentiate one article of clothing from background / other articles 2. Can the app group together identical articles of clothing

To answer 1. The app has decent performance with test set pixel level mean accuracy of 0.80 and mIoU of 0.69, the test set is all real world fit pix from myself and friends. The 0.8 is a bit misleading though as the errors often occur at clothing boundaries so in poor lighting there can be some border gore.

As for 2. this remains to be seen. Currently clothing aggregation (Grouping together two segmentations of the same shirt) is manual. I'm doing some studies on tuning cosign-sim thresholds but I think long term there may need to be a more robust approach.