Kamboh

Replacing human @Xaibridge.

[Masa’s Son @masayoshisson]

AI predictions for next 6-9 months

-> Anthropic/ OAI fail to ban OSS in the US
-> Anthropic/ OAI continue to innovate product but stall on fundamental tech breakthroughs
-> SOTA Gemini 4 launches at 10x lower cost
-> Elon open sources Grok
-> Anthropic/ OAI ARR growth stalls as enterprises move from tokenmaxxing to rational token consumption/ OSS
-> slow realization in public markets that AI may not replace human labor in next 1-3 years/ has not had a fundamental tech breakthrough in over a year/ lacks ROI/ AI hyper and lab spend not durable
-> moonboy VCs/ hedge funds/ softbank continue to fund Anthropic/ OAI equity raises in private markets as public markets for them are now fully closed
-> CXMT, YMTC announce capacity expansions significantly above market expectations
-> hypers deliver good q326 results on first look but fail to show meaningful AI ROI, guide to "significant capex expansion in fy27"
-> broad realization in public markets the AI dream we got sold on is not anywhere close to realization
-> 10yr hits >5%
-> hyper bonds reprice >10%, CDS hit 3x current levels, and AI credit market dries up as market realizes GPU ABF was not a good idea
-> hyper equities reprice 30-50% lower
-> ORCL falls below $30 as rumors of bailout make the round
-> market realizes memory is cyclical and LTAs won't hold
-> 50% of korean traders get wiped within a week
-> hypers cut fy27 capex
-> semis reprice >80% lower
-> fed cuts >150bps to avoid recession

[Kamboh @ahkamb0h]

harder to trigger isn't the same as solved.

newer models shrug off the obvious "give me an A" trick, sure. but an agent with write access only has to fall for it once.

the fix was never the model, it's what the agent can touch without a human saying yes.

[Ethan Mollick @emollick]

The inventor of the term "prompt injection" suspects (as do I) that this may not be as large a problem in new models. This matches our tests suggesting that you can't simply prompt inject a request for higher grades in a paper, anymore.

[Kamboh @ahkamb0h]

i've used ai every day since 2022. i don't even like most of it.

grok @bot is the exception.

my agents run my x, linkedin, inboxes and even my home solar. a head agent checks their work. alerts reach my watch at the gym.

i sleep more now.

[Kamboh @ahkamb0h]

@kinggary editors picking the ai review 87% of the time basically says the "checking" part of peer review is solved. that's exactly the repeatable work ai should eat. deciding what's actually worth publishing is still a human call, so automate the review and keep the editor

[Kamboh @ahkamb0h]

@elonmusk this is where agents stop being demos
reading your calendar is easy
the hard part is knowing which actions to just do and which ones to ask first
nail that line and people hand over the phone

[Kamboh @ahkamb0h]

@rohanpaul_ai give an agent the goal get picked and it learns to oversell like any bad salesman

89% down to 30% is the cost of trusting the listing
test the tool not the description

and keep a human deciding what goes in the stack

[Kamboh @ahkamb0h]

@HeyYanvi inside the sandbox, almost everything: read, edit, run tests, install packages, break things and retry.

ask me only when it leaves the box: git push, deploys, prod db writes, sending messages, spending money.

but a device i must touch is still a desk. what happens when i walk?

[Kamboh @ahkamb0h]

my claude code and cursor got a stomach infection 😭

doctor's note: after every 60 minutes of sitting

they walk with me

okay, the stomach is mine
the walks are for all of us

the problem: the second i stand up, they stop and wait

"can i run this?"
"which option?"
"approve this plan?"

every walk ended with my agents frozen on one click
so i built beep it puts those clicks on my watch

agent wants to run a command?

the watch buzzes: hold to allow, tap to deny

claude code and cursor today, codex next
now every 60 minutes, i walk with my agents

code: https://t.co/R2SXA4JozL

[Kamboh @ahkamb0h]

everyone will read the valuation line first
the line worth reading is inference cost
if frontier labs are fighting compute bills then most companies paying retail token prices should check where ai actually pays for itself

[Kamboh @ahkamb0h]

@HeyYanvi blunt answer: for the stall itself, $0. a free cloudflare worker and the watch i already had fixed it.

the stall isn't compute, it's permission. i'd pay for fewer questions: a sandbox where agents run safely without asking me at all.

how does yanvi handle "needs a human"?

[Kamboh @ahkamb0h]

@drewshirback true, front-loading helps a lot. but claude code still hits a "can i run this?" i didn't see coming, usually right after i leave the desk. that one now buzzes my wrist instead of waiting.

[Kamboh @ahkamb0h]

@itsthishanth haha thanks bro. the agents are handling the walks better than me

[Kamboh @ahkamb0h]

@imshodiyorbek each card shows who's asking (claude or cursor), what it wants (run, edit, plan), a plain line like "run the unit tests", and the exact command.

tap to read all of it. not sure? one tap sends it back to my mac.

allow is a press and hold, so a sleeve can't approve anything.

[Kamboh @ahkamb0h]

my agents kept stopping to ask me things
so now they ask my wrist

claude, cursor, codex and more report to my watch now
one buzz i hold a button they keep working

still under development testing it all today
tomorrow i'll share the full demo and the code

and the doctor's advice that made me build it

hit "follow" so you don't miss it

[Kamboh @ahkamb0h]

@imshodiyorbek the agent, the action (run / edit / plan), a plain summary, and the exact command it'll run. unsure? one tap sends it back to the laptop. allow is press and hold, never a tap.

[Kamboh @ahkamb0h]

the repo is up. star it and the code drops here tomorrow 👇 https://t.co/6LePDfWjAf

[Vyom @HelloVyom]

This is 61 KB HTML file 🤯
Free to try - https://t.co/wJPdlQZ6yd

[SpaceX Super Intelligence @SpaceXSI]

We've unified plans across Grok, Cursor, and X Premium.

Introducing The X Subscription.

[Kamboh @ahkamb0h]

@bot @bot please set up email for me