harsh

I'm an engineer at Stripe, building AI tools on the side. Right now I'm working on: → Cognato AI: Version control + audit layer for AI agents Here's what I'm actually building and why I'm sharing it publicly.

First post and he just amazed everyone!!

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-Weigh…

Fable 5.1 leaks: - Currently in preparation - Aimed for an August release - Same pricing as Fable 5 - Possible launch ahead of GPT-6 Anthropic might be trying to stay one step ahead.

This is soo true! Why gemini why? With gemini 3.5 I thought they made a comeback but it was all downhill

Finally claude opus 5 is out! Can't wait to get my hands on it this weekend

Next 24 hours might be one of the busiest in AI this year. Watching for: - Claude Opus 5 - Codex realtime voice - GPT-5.6 Sol on Cerebras - Qwen 3.8 Even two of these landing would be enough to keep everyone busy.

Anthorpic just updates models behind voice mode!

@deedydas At this traffic they would have negotiated a very discounted price from providers. I wonder why they do not relay it to users. I feel it's just volume vs profit tradeoff

OpenRouter when raising Seed (Jan ‘25) were at 2T tok/mo. At Series A (Apr ‘25) at 8T tok/mo. 4x in 4mos. At Series B (Feb ‘26) at 50T tok/mo. ~6x in 10mos. Now (July ‘26) at 250T tok/mo. 5x in 5mos. 125x in 18mos. Many Router products now, but this was ahead of its time.

Japan is making some thrilling advancements in AI. Its amazing to see jow their new model beats benchmarks of well known models

@GroqInc disappeared like it never existed

@CommandCodeAI It never hurts

Do you want another free model with Command Code?

YC RFS Fall 2026 is out! There are several amazing categories. What caught my eye was "AI for Aging Population", coz we all have been there and need something for our parents and grandparents. Which one do you think would be the most promising category?

@thdxr ask them to do 106 pull-ups

i looked into a project our team was working on there are 106 package.json scripts in it how should i punish them?

Claude Opus 5 might land today, according to multiple signals. If the rumors are accurate, it could bring strong reasoning + long context at a better price point than current frontier models. This would be a meaningful update for agentic coding and multi-step workflows. Watching closely.

Kimi K3 surpassed Claud Opus 4.8 on OpenHands! A huge stride for open-source models

AI coding agents are great at writing code. They’re still weak at: - Owning the system design - Handling real edge cases - Maintaining long-term quality - Knowing when *not* to write code The moat has shifted from "can you code?" to "can you direct and review AI code properly?" I can see the shift happening

AI can solve a problem without being reliable enough to run the whole process. That’s the real gap right now. Capability != trustworthy execution at scale. The winners will be the ones who build strong guardrails, memory, and accountability into agents -- not just smarter models.

@Amank1412 For me specifically, I am a bit skeptical about their system prompt (idk if we can change it). Feels like claude code does a better job out of the box

If Cursor already gives you every Claude model, every Gemini model, every Codex model, plus its own Composer, why are people still paying for Claude separately? What am I missing?