Ankit Shankar ретвитнул
Ankit Shankar
635 posts

Ankit Shankar
@thelastnode
Hacker; Palantir; Georgia Tech
San Francisco Присоединился Aralık 2010
492 Подписки1.8K Подписчики
Ankit Shankar ретвитнул
Ankit Shankar ретвитнул
Ankit Shankar ретвитнул

Former hobbits. If you have ever considered returning to Palantir, this is the week to do it. The world is demanding every last unit of creative energy we can muster. If you return, you will be on a plane day one and committing code that matters within hours of getting your laptop. From the foxhole to the factory floor the surface area to support something incredibly meaningful, immediately, is immense. I promise you nothing other than the sense of satisfaction that comes from the purpose and intensity of the most intense deployments you ever worked on. If you have been chasing that spark, come find it again. The shire is calling.
English

@rish404 That one I already knew about, but high quality, on-device STT is HUGE
English

@thelastnode You can also add an LLM node in the shortcut to make it 10x more powerful & practically do anything any TTS app can 😁
English

@JsonBasedman You should ask it to use it's context cleanup tool if you run into context issues. Sub-agents is architecturally supported, but might not be necessary for what you're doing!
English



New release of Rodney, my CLI tool for browser automation (designed for use by coding agents and with Showboat) - contributions from five people! simonwillison.net/2026/Feb/17/ro…

English

@LeeLeepenkman @bcherny Plan mode is one of the few places I actually need to make a decision (and it's highly levered), so generally feels like a good place for more options.
English

Yea, can do it automatically I guess.
I just need to have this full auto mode like these kind of cases where I'm making, I don't know, I just want to experiment try and see how much it can do on its own completely free.
It works sometimes like if you just making a new project to see if it can do something or if you've got like a really strong eval benchmark you can have it experiment in a new directory see if it does any better at algorithm development with some clear benchmark etc. just wanted to absolutely try its best without oversight
English

Probably the worst UX in Claude Code is that screen after it goes in plan mode. It does like, do you want to continue in dangerously skip permissions mode? It's like, obviously, yes. and with clearing context? obviously no. cc @bcherny
English

@bcherny @LeeLeepenkman Definitely prefer being able to clear context - that's the whole point of having a plan!
English

@LeeLeepenkman What’s bad about the screen — do you prefer not to have the options?
English
Ankit Shankar ретвитнул

@patrickc “The binding of quotation marks in English is fundamentally incorrect”, he added.
English
Ankit Shankar ретвитнул

I'm trying to figure out where to live next, and one big consideration is the climate. So naturally I made a tool that represents monthly average temperatures for cities as 3D rings so they can be compared more easily. Check it out: awjuliani.github.io/weather-explor…
English
Ankit Shankar ретвитнул

“Wow, this is a nice area” - in-laws going through Union Square
Incredible work @DanielLurie
English

@EthanReidMorro I use mise to install tools so that I can save my config in one spot, and it uses uv under the hood.
English

@EthanReidMorro It's just so fast and consistent. Beyond that, it also makes installing tools like moondream-station really easy.
Hard to go back to the old way after experiencing it.
English

Very cool - local image model that is very capable, now on macOS.
I prefer uv over pip: uvx moondream-station
moondream@moondreamai
Announcing MLX native Mac support for Moondream 3. Moondream running on Mac, Linux, Windows (free!): pip install moondream-station Details: moondream.ai/blog
English



