Jaski

Intelligence | Energy | Space
Ad Astra

[Jaski @Jas_Jaski]

The future shape of software

The thesis is that as models get better, protocols standardize, and agent runtimes become more capable, the center of gravity in software shifts away from standalone UI-heavy apps and toward services, automation, execution quality, and deep domain infrastructure.

Some of those services will still present themselves as branded apps. Some will be invoked through voice. Some will be reached through a marketplace. Some will be called through CLI-like execution surfaces or protocol connectors. Some will be embedded invisibly inside another company’s product.

The visible UI layer becomes thinner and more replaceable; the execution layer, the trust layer, and the domain-quality layer become more important.

[Jaski @Jas_Jaski]

Right now, I have 14 different AI agents running in parallel:
6 Codex threads
3 Grok chats
2 ChatGPT conversations
2 Cursor sessions
1 Claude Code
Zero issues managing them.

The problem isn't the AI. It's the friction of bouncing between them.
Built something really simple but insanely cool to fix it.
Launching next week via @CassiniRes

Open source. Fully local. Runs on a 16GB+ Mac.

[Jaski @Jas_Jaski]

@Hesamation Claude notoriously wastes your tokens to fake ‘impress’ you. Most tasks are much simpler to execute and don’t need a gazillion tokens.

[leo 🐾 @synthwavedd]

🚨 SCOOP(s):

- GPT-5.6 will be the final model in the 5.x series. GPT-6 is slated to launch in about a month, earlier than expected, and possibly even later this month
- GPT-6 will be based on a new, significantly larger pretrain (versus the ~4T 5.5/5.6 'Spud' base)
- There is lots of excitement at OpenAI over this new base, which they believe will be much better able to compete with both Fable 5 and upcoming 5.1, targeting a similar release window. OpenAI initially intended to continue with Spud through GPT-6, but decided against it
- On the topic of Fable 5.1, it is in the late stages of the pipeline at Anthropic and a release is expected "in the coming weeks"
- On the other side of the globe, DeepSeek are preparing for an imminent launch of V4 GA, which seems likely to be on par with or better than GLM-5.2, and have begun work on a new, larger model that will compete with the upcoming 2.7T MiniMax Pro

[Jaski @Jas_Jaski]

@___Harald___ @comma_ai Everyone should use Cursor & Codex.

[Jaski @Jas_Jaski]

I’ve been saying this for a while now. OpenAI’s lead on general tasks and long compute work is primarily due to abundance in compute and data.

[Deedy @deedydas]

“The dirty secret in AI is that everything is a data and an eval problem.

The best models have the best data and best internal benchmarks. The mid ones buy a lot of data, not the best, and hillclimb public benchmarks.

(you need a lot of compute too)”

– Stanford CS Professor

[Spencer Pratt @spencerpratt]

In a world of takers, be a maker. 🇺🇸🇺🇸🇺🇸

[Jaski @Jas_Jaski]

A dollar above $200/week is a waste. The sooner we realize it, the better.

This will also allow us to focus less on token maxing and more on getting deeper to solve intelligence and make it cheaper.

[Chamath Palihapitiya @chamath]

Tesla is one of the smartest, cracked and most advanced engineering companies in the world.

If they actually did this, then it is likely verifiably true that a dollar above $200/week is waste.

[Jaski @Jas_Jaski]

Just spoke to someone from Deepmind who 'secretly' believes AGI is a pipe dream. The irony of working at the frontier of AI and still failing to understand exponential scaling laws.

[Jaski @Jas_Jaski]

@deepansh_see Thanks @deepansh_see !

[Jaski @Jas_Jaski]

Last month I wrote 250,000+ words. But I typed almost none of it. 🤯

Juno is a voice layer for Mac.
Open source. Free. No account, no cloud.

The problem with voice tools has always been the tradeoff.
- The usable ones ship every word you say to a server.
- The private ones are not feature rich.

Juno doesn't make you choose. You press a key and talk - it writes the clean version of what you meant, makes the note or reminder, rewrites text you've highlighted, all in whatever app you're already in.

And it runs entirely on your Mac.
Your voice, your transcripts, your corrections - none of it ever leaves the machine. Nothing to leak, nothing to train on.

That's the hard part. Software this capable almost always needs the cloud - we got it running fully on-device, in real time, without losing any of the smarts. For teams, that means the sensitive stuff - client calls, code, contracts - never leaves the building. Also, finance heads love our $0 price tag.

Talking was always faster than typing.
Now it's finally private too.

We just went live on Product Hunt.

PH link is in the comments.
Would love your notes, questions and feedback

— Captured with Juno · Thu, Jun 18 at 3:44 PM

[Cassini Research @CassiniRes]

Top 10 out of 750+ launches on Vercel Day? We'll take it. 🤯
Yesterday, MindReader v1 locked in the #9 spot on Product Hunt with 125 upvotes, and the energy has been unreal.
If you haven't seen what we’re building at Cassini Research yet: MindReader v1 lets you paste in your pitch deck, ad copy, or sales emails, and literally simulates how a human brain responds to it - region by region.
No actual brain scanner required. We mapped this out using neuroscience models built on 35 years of fMRI research to give you immediate, actionable neuro-metrics. Stop guessing if your copy hits and start seeing how the brain actually processes it.
We’re also openly inviting academics to come stress-test our results - let’s see how far we can push this.
Check it out here: https://t.co/VtKavCekcm 🚀

[Ish @DecisionTree_gg]

How do you feel?
It's the oldest question in art and the newest one we can answer with technology.

MindReader takes any piece of content - a video, a post, a sales call - and simulates, second by second, how a brain responds. Where attention holds. Where effort spikes. What actually gets encoded into memory.

I've been running it with sales teams who want to know how a pitch landed in the buyer's head - deeper than the transcript, to see where attention dropped and where something stuck. We finally have* evals for conversation.

MindReader is built on Meta FAIR's TRIBE v2 + 35yrs of neuro research. The makers are engineers who've authored multiple academic papers and have been building open source for the last 4+ years.

Completely open source - I encourage you to tinker!

Run your latest social media post to see where attention dropped (or explore brain effort, gut reaction and 4 other neuro-dimensions).
The Product Hunt link is below

[Jaski @Jas_Jaski]

@__tosh unrelated, but just DM'd you about something nobody should hear but you [not spam, i promise]

[Jaski @Jas_Jaski]

@thisiskp_ Ecosystem is the way. Just DM'd you about something nobody should hear but you

[Jaski @Jas_Jaski]

@kevinwdavid AI is beautiful. Just DM'd you about voice

[Jaski @Jas_Jaski]

@zaczuo It is party time. Just DM'd you - mate!

[Jaski @Jas_Jaski]

@byalexai Hence I'm a big supporter of having local intelligence. Just DM'd you.

[Jaski @Jas_Jaski]

@fmerian (also, DM'd you)

[Jaski @Jas_Jaski]

@fmerian on it 🫡

[Carlos That Notices Things @QuetzalPhoenix]

The richest guy on Earth SHOULD be the guy making cutting edge cars and rockets instead of dudes who sell purses and perfumes or dudes who run investment firms or dudes who made Facebook.