Ajith Bhat (ಅಜಿತ್)

@bryncurran Awesome! Congratulations mate!

It's all worth it in the end. 🏅#LondonMarathon

OpenAI is shutting down text-embedding-3-small?!? I strongly believe that if you shut down a closed-source embedding model that you should open-source. Imaging the trillions of tokens that will no longer be queryable. cc @romainhuet

Congrats on the ship! Two ideas worth considering: 1. Invert the model — let users exclude topics (e.g. hide politics) rather than pick them in. Keeps the full following graph intact minus the noise. 2. Allow combining several topics into one tab with an optional filter, to cut the cognitive load of tab-hopping. A few things I'm running into already: - Same post shows up in both Tech and AI (duplicates across overlapping topics) - No indicator of which tabs I've already visited - Time-sensitive posts feel scattered across tabs and easy to miss

Ladies and gentlemen, today we're launching one of our biggest changes to 𝕏 Introducing Custom Timelines This feature allows you to pin a specific topic to your home tab. With support for over 75 topics, you can dive deep into your favorite niche on X. It's powered by Grok's understanding of every post with the algorithm's personalization—meaning every timeline is made just for you. And it works even better when it's a topic you already engage with. This was a huge undertaking across many months, so we're excited for you take it for a spin. We're giving early access to Premium subscribers on iOS (and Android coming very soon).

@denicmarko jobproof.dev

Share your websites.

Watching the Claude Opus 4.7 launch reactions split down the middle today made me want to write down something I've been thinking about for a while — why this exact fight happens every single model launch, and what it's actually about.

@thekitze Happy birthday @thekitze !

don’t even ask where I’m gonna eat my birthday meal

Spotty signal, background noise, rushed sessions — @getjobproof handles it all a lot better now. Big reliability push this week. jobproof.dev

@PKodmad Hey Duggee, Bluey, and a couple of sticker/colouring activity books. Best of luck. 😂

Going on a long haul flight tomorrow. Time to bust out the iPad for my 3 yr old. She hasn’t used it in more than 6 months. What can I download on it that can engage her and my 2 yr old on a 14 hour day flight?

We spent a decade learning that personalised search creates filter bubbles. Now AI assistants come with persistent memory, quietly shaping responses around what they think you believe. They’re not hiding answers. They’re framing them. Which might be worse.

@thsottiaux @reach_vb Do quota resets create a thundering herd? If everyone’s weekly quota renews at once or gets globally reset after an incident, I’d expect a spike right at the boundary. Do you mitigate with staggered resets / jitter, or is load smoothing enough?

@PKodmad "Works on my machine" 😅 But seriously, just pushed a fix. Should work better now. Give it another try? 🙏

@AjithBhat The mic permission is buggy. I couldn’t get it working

Built an app that nags my kid to eat faster so I don't have to. It loops my voice forever. My throat is free. My daughter is not amused 😂 bega-bega.vercel.app ಬೇಗ ಬೇಗ

A few things that help when you're flying blind on vibe-coded Swift: 1. Ask Claude to add logging/instrumentation first, reproduce the bug, then diagnose from the actual output — it reasons much better from evidence than description 2. Ask it to web search the specific APIs it's using — SwiftUI has known performance gotchas per version that are easy to miss 3. Ask it to isolate the slow path into a minimal reproducible case — vibe-coded apps tend to have tangled state, and isolation usually reveals the real culprit

I’m hitting my first set of persistent performance bugs with a completely vibe coded app. It’s in swift ui so my understanding of the code nuance is very limited. Each time i describe the bug to Claude, it comes up with a plausible plan, yet the bug persists. Codex broke critical features when given the same task. I wonder how to navigate this

Shoutout to @gilbarbara for react-joyride. Added a product tour to our app @getjobproof and it just worked. Great library, great DX. Thank you for maintaining it 🙌

Great interview @bcherny @lennysan — loved the conversation on Claude Code. Been using it daily as a solo founder shipping AI products. A few things that would take it to the next level: 1. Smarter planning for agent-driven workflows. Two issues here. First, planning mode creates Phase 1-5 roadmaps that defer features weeks out — which makes sense for human teams, but when a coding agent does the work, you can ship the whole thing in a day or two. Deferring what could be built in 20 mins to Phase 5 is counterproductive. Second, the planner makes simplifying assumptions in early phases to move fast — things like project structure, scripts, config. Then when later phases need something different, you’re reworking what was already built. With a human team you’d call that a refactor sprint. With an agent that can do it all in a sitting, there’s no reason not to build it right from the start. It’d be powerful if the planner could ask whether humans or agents will execute, think through the full picture upfront, and prioritise by dependency order rather than time-boxing. 2. /summarize command. Long sessions build up so much context — decisions, dead ends, what’s left to do. When the session ends or you hand off to a new instance, that’s all gone. A simple /summarize to auto-generate a clean session snapshot would make restarts and handoffs seamless. Small feature, massive quality of life. 3. Delegate to an Extended Thinking sub-agent for deeper research. Right now if I need proper analysis, I’m bouncing between web search and the TUI manually. It’d be great to delegate from the plan — pass context, files, and a question to an Extended Thinking sub-agent that does the multi-step research and comes back with a synthesised answer. All three come from the same place: Claude Code is already brilliant. These would just close the gap between how it’s designed and how power users actually work with it. @AnthropicAI