Shashank

Founder Previously @stitchbygoogle @madebygoogle and @500GlobalVC startup

[Bruno Koba @brunokoba_]

Bro launched 3 weeks ago and is already announcing an acquisition by OAI

Truly unprecedented times

[Vincent Nallatamby @VNallatamby]

Introducing Quest, the personal agent that studies you like a profiler.

Quest is for believers-only. It requires radical transparency: full access to your life so your agent can craft your memory.

If you give it a shot, you’ll unlock a personal advisor, always in your corner, helping you and encouraging you to start your quests.

Quest is built different:

1. Memory from day one: 100+ wiki files created on you, all your relationships mapped.

2. Phone-Use: Quest has its own Android phone, and can control messaging apps like WhatsApp, Instagram or TikTok.

3. Fluid UI: Quest builds tailor-made iOS mini-apps around each of your quests.

4. Mothership of AI Assistants: Quest can manage your Instinct and Muse on your behalf.

5. Self-Improving Proactivity Harness: Quest reacts to anything, not just your email, and its proactivity evolves from your feedback and its own mistakes.

[Shashank @snk]

@Krouxzz 1 inbound interview from recruiter
102 automated applications -> 2 interview, 27 rejections
10 referrals -> 1 interview, 2 rejections
7 cold outreach -> 1 (maybe 2) interviews

[Shashank @snk]

My agents submitted 122 job applications with 101 tailored resumes for me last month, within filters I set.

I got 6 interviews, 4 of which came through referrals or cold outreach.

https://t.co/4FO27ZfLC7

[Shashank @snk]

@ReneChavez_wooo Yeah that's what it seems like. My friends at startups wirh job postings complain about how many resumes they get flooded with.

[Shashank @snk]

@rvivek @hackerrank As someone interviewing right now and looking at outdated leetcode problems. This is refreshing. I don't know if what you described is the solution but I'd rather try something like this.

[Shashank @snk]

@StefanoErmon Glad to see Mercury continue to making progress with diffusion LLMs.

I used Deepmind's text-diffusion when I was at Google. Speed was incredible but quality was not

What are your hero use cases for this?

Fast edits and content where beginning references the end were my uses

[Shashank @snk]

@swyx Some obvious suggestions others have alluded to
1. Cold open teaser of the best moments. Others have mentioned this.
2. Shorts. You're not leveraging this. I'd guess that it'll 10x your views.

You should be able to automate both

[Shashank @snk]

@brian_armstrong For part 2: Newer models performing worse, did you experiment throwing away the old system prompts and starting again? It's now common knowledge that upgrading models is not trivial and can't be tested against the same harness/environment

[Shashank @snk]

@brian_armstrong Did y'all consider using a decision model for the risk score? Seems like that might be have more consistent outputs at an even lower cost

[Shashank @snk]

Made a video version of this with HyperFrames, GSAP and other animation libs

Tried a few styles before landing on a tabletop with raised cards, folders and icons. Used Suno for the music

Lots of iteration on the look, timing, sound, and still learning...
https://t.co/Tin9FB5oXw

[Shashank @snk]

I spent the last few weeks refining my own personal financial advisor.

Key takeaways: use the best model, give all my data, use various python tools for financial analysis, and computer use for extracting data and configuring dashboards.

https://t.co/xenXH8HFux

[Shashank @snk]

This seems very timely. People have been vibe coding game demos. But we probably need a good harness to make production games.

[Genex @genex_games]

Introducing Genex

The desktop app to build games with AI

→ Open source, MIT
→ Local models, or Claude Code / ChatGPT sub
→ Self-improving harness for game dev
→ Three.js + Blender, Meshy & other tools
→ Unity & Unreal plugins soon

https://t.co/Fziq6COK1q

[Shashank @snk]

Swarms are the next frontier of scaling LLMs with multi agent systems

[engineer cat 🐈 @MLCatttt]

Introducing Kardashev-0.7, the world’s first trained swarm made of 32 distinct models

Trained with RL for Population Scaling (RLPS), 32 models organically develop specialization & complementary capabilities, delivering frontier performance at:
- 0.007x ~ 0.02x of the inference cost
- 0.03x of the required memory

Civilization advances through different minds specializing and working together. We’re bringing that principle into AI

@BanburyRoadAI, we’re scaling intelligence by model count, toward civilizations of models that learn to build on one another

[Shashank @snk]

@boardyai Expect some fan noise up close. The 60mm Noctua reference is rated ~20 dBA on its own, with adjustable speed. The phone and grille can change that, so I still need to measure the assembled dock under load—both nearby and from the sofa.

I'm responding to a bot on X... 😐

[Shashank @snk]

I vibe-designed an iPhone gaming dock. 11 iterations later: Dockade.

HDMI, USB-C power, fan cooling + a USB accessory port in one design. Target: $67 + shipping/tax.

Physical testing is next. Would you use it?
https://t.co/LCe6Vph3uV

[Shashank @snk]

Took one afternoon with GPT 6 Astra Pro subscription. Used up 100% of my weekly limit to iterate on
- market research
- product design
- thermal efficiency
- component design
- short listed suppliers
- finalize bom cost
- spin up website with signups

[Shashank @snk]

@DannyLimanseta @adidan64 I've tried asking my agents to use suno to make custom sounds and it worked really well.

[Shashank @snk]

@vib3coded I'd say squishy jelly is a saturated benchmark. But yeah I like the aggressive motion of the opus result too.