Amrish Kaushik

enterprise ai | payments

[Amrish Kaushik @amrishkaushik]

@levie It’s equal parts model eval and compliance/governance imo. As specialized vertical models gain maturity/function; they will become more stochastic / differentiated, needing swarm routing (think L0/L1/L2 cascade in human workflow)

[Marko Dinic @mdinic]

There’s no AI pricing war. There’s Google with its own silicon, and there’s everyone else renting the factory.

At I/O 2026, Google claimed enterprises could save $1B+ a year by moving 80% of workloads off OpenAI and Anthropic onto Gemini 3.5 Flash. Sounds aggressive until you realize Google is the only one who can afford this. OpenAI and Anthropic rent their compute. Google built it.

Enterprises were already drowning in token bills before Google walked on stage. Uber burned through its entire 2026 Claude Code and Cursor budget in four months. Adoption hit 84% of 5,000 engineers. Heaviest users ran $500 to $2,000 a month. GitHub moves Copilot to token-based billing June 1.

If you’re buying AI for an enterprise, stop modeling it like SaaS. No annual flat-rate lock-ins. Treat tokens like cloud spend with caps and per-team budgets. Build for multi-model routing, because providers without their own silicon will keep getting squeezed.

Owning the silicon is the only real moat in AI right now. Everything else is margin compression dressed up as innovation.

[Amrish Kaushik @amrishkaushik]

Anyone working on GoBD or SEC 17a-4 compliant Enterprise AI platforms?

[Jason Shuman @JasonrShuman]

We’re entering the Data Disintermediation Era.

The most valuable startups are not replacing systems of record. They are intercepting data before it ever reaches them and owning the workflows that follow.

The last two years were defined by AI voice and scribe companies. Not because transcription was novel, but because they captured high-value data upstream and used it to trigger billing, customer acquisition, and other revenue-critical workflows.

The next phase will be even more powerful.

The next two years will be defined by physical-world data disintermediation. Fixed or mobile/portable hardware will capture proprietary data at the source, where switching costs are real and ROI is directly measured against labor, error rates, revenue boosting activities, and compliance risk.

Software can be replaced by the core SoR rolling out competitive products, but Hardware-anchored data pipelines will be challenging to reach to for incumbents.

[Amrish Kaushik @amrishkaushik]

@_equityXplorer @TuckerCarlson @ScorpionFund @Kir_Kahlon IMO, the key thing about $TMDX technology is that warm perfusion significantly extends the *time* the donor organ remains viable i.e. no time pressure or need to hurry any procedures.

[Andrew Ng @AndrewYNg]

Agentic Document Extraction now supports field extraction! Many doc extraction use cases extract specific fields from forms and other structured documents. You can now input a picture or PDF of an invoice, request the vendor name, item list, and prices, and get back the extracted fields. Or input a medical form and specify a schema to extract patient name, patient ID, insurance number, etc.

One cool feature: If you don't feel like writing a schema (json specification of what fields to extract) yourself, upload one sample document and write a natural language prompt saying what you want, and we automatically generate a schema for you.

See the video for details!

[Amrish Kaushik @amrishkaushik]

@gokulr ERP platforms have the most proprietary data though; imho, it will eventually come down to who has the better data to impact their specific business/use case; still couple of iterations away wrt time series indexed etc.

[Amrish Kaushik @amrishkaushik]

@gokulr 💯 agree - a mix of product and a strong understanding of business process is needed. Otherwise it’s a hammer looking for a nail.
Eg Healthcare - pre-auth automation, medical billing/coding, nurse co-pilots, etc

[Sully @SullyOmarr]

Gemini 1.5 pro is STILL under hyped

I uploaded an entire codebase directly from github, AND all of the issues (@vercel ai sdk,)

Not only was it able to understand the entire codebase, it identified the most urgent issue, and IMPLEMENTED a fix.

This changes everything

[Amrish Kaushik @amrishkaushik]

@balajis @balajis one difference between Blue Democracy and Indian Democracy is that at the state level, there are more “tribes” as you put it, with more real power, own agenda and independence vs center/federal.

[Amrish Kaushik @amrishkaushik]

@HarryStebbings IMO, Deconstructing legacy workflows into fintech point solutions focused on simplicity and ease of use..

[Amrish Kaushik @amrishkaushik]

@elonmusk As a parent, I could get behind this legit exercise.. Let’s turbo-charge the PE period with metaverse incentives and see a whole generation of uber-fit kids grow up 😂.. there are only about 70,000 middle schools in the US.. what say @elonmusk ?

[Dev @beef_wagon]

@VitalikButerin 1. Vitalik announces ETH 2.0 is 50% done.

2. Vitalik says ETH "could probably" be done in 6 years.

3. Vitalik posts this poll "so guys, if ETH DOESNT prevail, who else we goin with?"

4. Eth community votes up BTC/SOL out of spite for its only real competitor, #Cardano.

[Amrish Kaushik @amrishkaushik]

@IOHK_Charles @WeiserEveryDay More like Inflation holds you to account..

[Charles Hoskinson @IOHK_Charles]

@WeiserEveryDay One does not simply account for inflation

[Mashable @mashable]

This augmented reality room is so trippy

[Santiago @svpino]

Free 2021 machine learning course from Harvard.

Topics:

• Basics of machine learning
• Popular algorithms
• Recommendation systems
• Cross-validation and Regularization

https://t.co/g2aINjCs6b

[Raoul Pal @RaoulGMI]

Something to get your head around:

Head Line:

A major asset class crashed 42% in 14 days, wiping out $1.02trn in value in an orgy of liquidation of people up to 100 x levered, with very low regulation. Many tokens fell up to 70%, including unregulated lending and borrowing biz.

[Amrish Kaushik @amrishkaushik]

@Austen IMO, this is not so much about control over one's emotions, but rather control over your response.
It's the ability to choose how/when to respond in the presence of stimulus/emotion/empathy/anger/disgust.