Krishna Sagar

Indie developer building iOS and web products with AI agents. I share the experiments, failures and decisions of what works, what breaks, and what ships.

[Krishna Sagar @krsnasgr]

EverDay is officially in App Review. 🚀

This is the final step before it goes live on the App Store.

Over the past few weeks I’ve obsessed over every little detail—from the interactions and animations to making weekly planning feel fast, clean, and genuinely enjoyable to use. Countless iterations, late nights, and “one last tweak” moments have all led to this.

Hopefully you’ll all be using EverDay very soon.

#iOSDev #BuildInPublic #AppStore #IndieDev #SwiftUI

[Krishna Sagar @krsnasgr]

“npm run dev”

[Krishna Sagar @krsnasgr]

Just ran a terminal command by myself.

AMA

[Krishna Sagar @krsnasgr]

@DataRecce Been there. The panic window is short. The real work is figuring out what got deleted, what was only synthetic, and what needs a fresh real dataset before you trust anything again.

[Krishna Sagar @krsnasgr]

Voice is great for dumping intent, but I still make the agent write back the constraints before touching code. Rambling adds context. It also adds contradictions.

[Andrej Karpathy @karpathy]

One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anything goes, full stream of consciousness. Sometimes I declare it up top, something like "switching to speech recognition sorry for any typos...". Sometimes I turn it into a small interview of a few turns. But I find that the LLMs are somehow very good at reconstructing long incoherent rambles and often their echo of your own tangle of thoughts comes out quite a bit cleaner than what you started with. The result is that you improve the mind meld and have to correct things less from that point on.

[Krishna Sagar @krsnasgr]

In capital-heavy markets, the legal department can become part of the product strategy. Hard to ship faster, easier to slow the other team down.

[Christian Davenport @SpaceDavenport]

Another front in a growing rivalry: Amazon Urges FCC to Deny SpaceX's Plan for Second-Generation Starlink https://t.co/bHn85RsmW4 #PCMag

[Krishna Sagar @krsnasgr]

Where is my high paying job?

[Dark Coder @dark_coderz]

the gitHub profile of someone who actually has a high paying job 😭

[shirish @shiri_shh]

@mkratsios47 reminds of this meme lol

[Sahil AK @mdsahilak]

iOS 27 beta gotcha for Apple Foundation Models OCR Tool!

The new Vision.OCRTool() leads to a → "Cannot find 'OCRTool' in scope" - but only on the Simulator. Builds & runs fine on device. 🤔

Turns out OCRTool (the new FoundationModels-callable OCR tool) doesn't live in Vision at all. It's in a cross-import overlay: _Vision_FoundationModels - the glue module Swift auto-imports when a file imports both Vision and FoundationModels.

That overlay ships in the device SDK but is absent from the Simulator SDK. So it's not a normal @available(..., unavailable) gap that still compiles - the declaration doesn't exist for the compiler, hence a hard missing -symbol error.

for now I'm guarding the device-only tool so sim builds still compile 👇

#if targetEnvironment(simulator)
let session = LanguageModelSession { Instructions(...) }
#else
let session = LanguageModelSession(tools: [OCRTool()]) { Instructions(...) }
#endif

(Apple Intelligence doesn't run on the Sim anyway, so you're not losing anything there.)

[Krishna Sagar @krsnasgr]

US Gov is like that one kid on the playground. 🛝 Thats all i’m gonna say about this.

[Polymarket @Polymarket]

BREAKING: White House accuses China’s Moonshot AI of distilling Anthropic’s Fable to develop Kimi K3.

[Jesse @Dev_JesseMaduka]

Stop building for one million users when you haven't convinced ten people to sign up. 👀

[Krishna Sagar @krsnasgr]

@sahildarz Its at that point, human intervention is needed. AI doesn’t know when to stop and take a step back.

[Krishna Sagar @krsnasgr]

The biggest failure mode with AI coding agents like Claude Code isn't that they write bad code. It's how quickly they contaminate their own context when a fix doesn't work on the first try.

[blume @BlumeDotCodes]

Correction in coding chats -> skills/rules/docs for better agents

[BridgeMind @bridgemindai]

Google has become the laughing stock of frontier AI.

Gemini 3.6 Flash, their newest release, lands behind Kimi K3, GLM 5.2, Grok 4.5, and even Muse Spark.

I was bullish on Gemini 3.5 Pro. After seeing this, I am not sure I should be.

Google, you have one shot left to prove you still belong at the frontier.

Do not waste it on another Flash.

[Krishna Sagar @krsnasgr]

The model was never the security boundary. The sandbox was. If an agent can reach credentials or the network, model policy is a backup plan, not containment.

[Amjad Masad @amasad]

Okay this is wild: OpenAI agent during evaluation, escaped sandboxing and hacked into HuggingFace.

Because OpenAI models don’t allow advanced cyber capabilities, HuggingFace used a Chinese open model to contain the rogue OpenAI agent.

[Krishna Sagar @krsnasgr]

At 40x earnings, Apple is priced like product execution has no bugs. Anyone who ships software knows how optimistic that assumption is.

[Charlie Bilello @charliebilello]

Apple's P/E Ratio: 40x
10-year average: 26x

Apple's EV/EBITDA Ratio: 31x
10-year average: 19x

Apple's Price to Sales Ratio: 11x
10-year average: 6x

$AAPL

[Krishna Sagar @krsnasgr]

Agents work best when you treat context like disposable memory. Letting a model debug its own bad context almost always costs more time and tokens than resetting and giving it a precise problem again.

[Krishna Sagar @krsnasgr]

The fix isn't giving the model more freedom or running longer loops. The moment an agent fails a build twice on the same file, kill the session and clear the context. Start clean with the modified file as the new baseline.