Hacker News: amanhijhttps://news.ycombinator.com/user?id=amanhijHacker News RSShttps://hnrss.org/hnrss v2.1.1Sat, 16 May 2026 10:54:12 +0000<![CDATA[New comment by amanhij in "Cursor admits its new coding model was built on top of Moonshot AI's Kimi"]]>I made a post previously saying that the release of Composer was so odd. It takes lots of dedication and compute power to achieve frontier level, and they suddenly have a benchmark score saying composer is frontier. I love all the innovation that is happening I just hate when actual good product gets shadowed by marketing mongers.

]]>Mon, 23 Mar 2026 02:45:48 +0000https://news.ycombinator.com/item?id=47484952amanhijhttps://news.ycombinator.com/item?id=47484952https://news.ycombinator.com/item?id=47484952<![CDATA[New comment by amanhij in "LLMs can't reason and never will"]]>So you're in the Yann Lecun bucket then? I agree his approach is great, however the current LLM approach is far from over we haven't even touched the upper end of it's abilities.

]]>Mon, 23 Mar 2026 02:42:09 +0000https://news.ycombinator.com/item?id=47484923amanhijhttps://news.ycombinator.com/item?id=47484923https://news.ycombinator.com/item?id=47484923<![CDATA[Why do we need apps like cursor?]]>I'm a Claude code and codex maxi, use both on the highest plan on my terminal and the web interface for heavy research and sometimes design. Now the whole moat of apps like cursor is multi agent orchestration if I understand correctly? And no cursor new coding model is not a moat there is no way on earth that they can compete with anthr/openai. For me I just launch claude code and tell it to run codex as one of it's sub-agents and split the work with sub-agents like half opus's and half 5.4's. What are y'all thoughts? I think any wrapper and orchestrator layer is going to 0

Comments URL: https://news.ycombinator.com/item?id=47474950

Points: 2

# Comments: 1

]]>Sun, 22 Mar 2026 06:18:52 +0000https://news.ycombinator.com/item?id=47474950amanhijhttps://news.ycombinator.com/item?id=47474950https://news.ycombinator.com/item?id=47474950<![CDATA[New comment by amanhij in "Delve – Fake Compliance as a Service"]]>Yeah that's a wild billboard lmao. btw 99% of MIT people are no different then the rest they just worked hard or paid hefty amounts, I have lots of friends that went there. Nonetheless the 1% are geniuses. Also saying MIT dropouts instantly makes your story credible, it's a funny concept. I'm starting to feel like an MIT dropout these days.

]]>Sun, 22 Mar 2026 02:31:46 +0000https://news.ycombinator.com/item?id=47473879amanhijhttps://news.ycombinator.com/item?id=47473879https://news.ycombinator.com/item?id=47473879<![CDATA[New comment by amanhij in "Delve – Fake Compliance as a Service"]]>I get where you are coming from, but still claiming "AI native" shouldn't change anything when it comes to due diligence. I agree tho the 'AI stamp" is letting a lot of things through.

]]>Sun, 22 Mar 2026 02:27:36 +0000https://news.ycombinator.com/item?id=47473859amanhijhttps://news.ycombinator.com/item?id=47473859https://news.ycombinator.com/item?id=47473859<![CDATA[New comment by amanhij in "Delve – Fake Compliance as a Service"]]>Lots of companies affected this, what blows my mind is when VC's were funding this how come no due-diligence was done on something as important as compliance. who even tries to scam on compliance like it's a known way to get caught.

]]>Sat, 21 Mar 2026 04:46:01 +0000https://news.ycombinator.com/item?id=47464058amanhijhttps://news.ycombinator.com/item?id=47464058https://news.ycombinator.com/item?id=47464058<![CDATA[Why do we need apps like cursor?]]>I'm a Claude code and codex maxi, use both on the highest plan on my terminal and the web interface for heavy research and sometimes design. Now the whole moat of apps like cursor is multi agent orchestration if I understand correctly? And no cursor new coding model is not a moat there is no way on earth that they can compete with anthr/openai. For me I just launch claude code and tell it to run codex as one of it's sub-agents and split the work with sub-agents like half opus's and half 5.4's. What are y'all thoughts? I think any wrapper and orchestrator layer is going to 0

Comments URL: https://news.ycombinator.com/item?id=47464045

Points: 2

# Comments: 3

]]>Sat, 21 Mar 2026 04:42:51 +0000https://news.ycombinator.com/item?id=47464045amanhijhttps://news.ycombinator.com/item?id=47464045https://news.ycombinator.com/item?id=47464045<![CDATA[New comment by amanhij in "Show HN: I solved Claude Code's context drift with persistent Markdown files"]]>any benchmarks?

]]>Fri, 20 Mar 2026 08:57:59 +0000https://news.ycombinator.com/item?id=47452118amanhijhttps://news.ycombinator.com/item?id=47452118https://news.ycombinator.com/item?id=47452118