frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Blanets and Weird Worlds

https://www.antipope.org/charlie/blog-static/2026/10/bizarre-blanets-and-other-weir.html
1•pavel_lishin•57s ago•0 comments

Debcraft at v1.0

https://optimizedbyotto.com/post/debcraft-1-0-released/
1•ottoke•1m ago•1 comments

Echo-3 and the SpAItial App

https://spaitial.ai/blog/echo-3-release
1•frozenseven•2m ago•0 comments

Doctor removes prostate from 35 miles away in first telerobotic surgery in US

https://www.theguardian.com/us-news/2026/oct/27/first-telerobotic-surgery
1•bookofjoe•2m ago•0 comments

I benchmarked 4 web search APIs for AI agents on 30 live queries

https://github.com/psinglap/websearch-api-eval
1•pallavisingla•3m ago•0 comments

A critical look at self-distillation for reasoning

https://jasonkena.github.io/blog/posts/does_self_distillation_improve_reasoning/
2•RevoGen•8m ago•1 comments

The Slopocalypse Survival Guide

https://alicegg.tech/2026/10/27/slopocalypse
3•zer0tonin•8m ago•0 comments

Cloud Rats and Software Roaches

https://idlewords.com/2026/10/cloud_rats_and_software_roaches.htm
2•pavel_lishin•9m ago•0 comments

Intel: Panther Lake

https://en.wikipedia.org/wiki/Panther_Lake_(microprocessor)
2•tosh•9m ago•0 comments

Claude vs. DeepSeek vs. GLM: what coding agents pick

https://whatagentspick.com/claude-vs-deepseek-vs-glm/
2•nsiradze•9m ago•0 comments

Satya Nadella Rethinks Whether More Powerful AI Is Safer

https://www.businessinsider.com/satya-nadella-ai-alignment-changed-mind-different-problem-scales-...
3•hirodxb•10m ago•1 comments

Wanted: Robot-Dog Walkers. $20/Hour

https://www.nytimes.com/2026/10/27/technology/robot-dog-walkers.html
2•beansovertoast•10m ago•0 comments

Show HN: Search2o Simplifies AI Agents

https://search2o.com/
2•ramvenkat11•10m ago•0 comments

Show HN: Agent-relay for multi-vendor coding sessions

https://github.com/saranyan/agent-relay
2•svts•13m ago•0 comments

Bunkering Down

https://interconnected.org/home/2026/10/24/bunker
3•tonyedgecombe•14m ago•0 comments

Almost isn't good enough: Introducing VybBench, our Design Fidelity Benchmark

https://cocolabs.ai/blog/introducing-vybbench
2•dzats•14m ago•1 comments

Show HN: Qev – quake and local decision models

https://twitter.com/shinyrobject/status/2115124326194393595
2•kernel_sanders•14m ago•0 comments

iPhone Duo Breaks After 75,001 Folds Considerably Lower Than Samsung Foldables

https://gizmodo.com/iphone-duo-folding-test-worse-than-samsung-foldables-2000831131
2•HieronymusBosch•15m ago•0 comments

Show HN: Builder Lens (Paxel Inspired)

https://github.com/saranyan/builder-lens
1•svts•15m ago•0 comments

Ukrainian launch test military-purpose communications satellite into orbit

https://www.pravda.com.ua/eng/news/2026/10/27/8059951/
3•uxhacker•15m ago•0 comments

Prose and Thorns – romantasy book discovery with AI-synthesized quality scores

https://proseandthorns.com
1•mattbaum•16m ago•0 comments

"Boy Born on Tuesday" a Thornier Problem Than the Monte Hall Problem?

https://win-vector.com/2026/10/27/boy-born-on-tuesday-a-thornier-problem-than-the-monte-hall-prob...
2•jmount•17m ago•0 comments

Agentic AI Foundation Adds Four New Open Agent Projects

https://techstrong.ai/building-with-ai/featured/agentic-ai-foundation-adds-four-new-open-agent-pr...
1•CrankyBear•19m ago•0 comments

Arch-Specific SIMD in Go

https://go.dev/blog/archsimd
2•tosh•19m ago•0 comments

Show HN: AgenticBench, 9 security tests run on 25 AI coding agents

https://agenticbench.org/
1•pantaleone•20m ago•0 comments

If a Function Is Mediocre, I'd Check Its Leader First [video]

https://www.youtube.com/watch?v=QCWAMWQ8A6A
1•peter_d_sherman•20m ago•0 comments

Two-thirds of car sales in China are now electric vehicles

https://www.carbonbrief.org/chart-a-record-two-thirds-of-car-sales-in-china-are-now-electric-vehi...
1•toomuchtodo•20m ago•1 comments

The Slack and Salesforce post-acquisition culture clash

https://buildingslack.com/salesforce-part-two/
1•anoojb•23m ago•0 comments

Ghostty 1.4

https://twitter.com/mitchellh/status/2115124583170974193
3•paaloeye•23m ago•0 comments

Introducing LFM3: Every Modality for Specialized Intelligence

https://www.liquid.ai/blog/lfm3
10•Philpax•23m ago•3 comments
Open in hackernews

Ask HN: If building an IDE from scratch, what features would you add?

2•rajasimon•1y ago
Many IDEs feel bloated or outdated. If you could design one from scratch, what would you keep, drop, or reinvent?

Comments

T0V1•1y ago
I'd keep all the features from neovim, remove all the features from every other IDE.
joegibbs•1y ago
Native vector search would be great. I was just thinking today how irritating it is to try to find some specific component in a very large codebase without knowing exactly what it’s called already.
SvenL•1y ago
Great debugging capabilities, time travel debugging, data break points, attach to running processes, remote debugging etc.

I think the most important feature for an IDE is debugging - at least this is the feature which separates an ide from just a Text Editor

pancsta•1y ago
You mean like Zed? Check their roadmap.