based on what we're seeing internally, scaling pretraining is a necessity if you want to operate near the frontier (this doesn't always translate to larger models though!)
The trend is larger and larger models
Kimi is now at 3T, next year we will probably be at 10T, and by 2030 probably 100T
7
2
40
2,383
You can just go to LCR.net and give your agents access to a fleet of 10s of thousands of browsers that have the ability to route requests through millions of real residential IP addresses.
notice..
> Delta is blocking AI agents from booking flights
> JPMorgan Chase and Charles Schwab block agents from accessing accounts
> Yelp requires paid data licensing for agents
> Amazon has blocked Grok bot and other shopping agents.
the internet is starting to block AI agents that were supposed to do everything for us
7
16
104
25,380
A very important milestone for Grass, and one I’m particularly proud of.
There’s a much bigger story behind where this is going, but for now: huge credit to the team that built it.
There’s a much bigger story behind where this is going, but for now: huge credit to the team that built it.
Today, we’re launching our first Live Context Retrieval tool for agents.
Introducing the Grass Contents API — an API that turns any public URL into clean, model-ready HTML, Markdown, or text.
42
21
133
7,047
My hot take is that it’ll probably take a while for personal agents to begin executing everyday purchases on behalf of people on a meaningful scale, and are more likely to act primarily as recommendation engines / price trackers for a while.
Main reason being many people love the act of shopping itself (actually you can probably safely argue that the majority of consumer use is scrolling social media & scrolling shopping websites, and consumers are addicted to both).
That said, if a major use case of “agents” for consumers ends up being shopping advice, what matters a lot less than the execution medium (whether the person or the agent clicks “buy”) is the fact that the agent is accurately reading the real-time state of each webpage. Nobody wants to use an agent for shopping if it’s recommending outdated prices or stock.
In other words, if agents suck at public web access, they will quickly become irrelevant for most consumer use cases. It turns out that the heaviest antibot occurs on e-commerce and travel websites, and will only show accurate data to consumer device fingerprints / IP addresses…
Main reason being many people love the act of shopping itself (actually you can probably safely argue that the majority of consumer use is scrolling social media & scrolling shopping websites, and consumers are addicted to both).
That said, if a major use case of “agents” for consumers ends up being shopping advice, what matters a lot less than the execution medium (whether the person or the agent clicks “buy”) is the fact that the agent is accurately reading the real-time state of each webpage. Nobody wants to use an agent for shopping if it’s recommending outdated prices or stock.
In other words, if agents suck at public web access, they will quickly become irrelevant for most consumer use cases. It turns out that the heaviest antibot occurs on e-commerce and travel websites, and will only show accurate data to consumer device fingerprints / IP addresses…
another MASSIVE idea:
it’s become glaringly obvious that personal ai agents won’t monetize users by charging $20/month.
instead they’ll monetize on the backend by taking a cut of the spend they drive to merchants.
which means every agent is eventually going to want the same thing:
discounted pricing + rev share from airlines, hotels, restaurants and every other merchant they funnel spend to.
but it makes absolutely no sense for every agent company to individually negotiate bulk pricing with thousands of merchants.
so build one network that aggregates all of the spend and:
> negotiates discounted rates
> secures referral / rev share economics
> wraps thousands of merchants into one API
basically a giant affiliate network for ai agents, except the purchasing power of every agent is pooled together.
this concept is nothing new either.
companies like Amex have been doing versions of this forever: aggregate huge amounts of demand, negotiate better rates and monetize the spread.
but now imagine that across every vertical of commerce and making it available to every ai agent through one API.
whoever builds this will end up sitting on a Visa-like network for agentic commerce.
2
5
66
8,531
yes actually, and more on this soon
@0xdrej appreciate the transparency here! if i was listening to this interview correctly - kinda sounds like you guys are starting to gather data for robotics? x.com/i/broadcasts/1…
16
5
49
7,479
Grass has been criticized in the past for its comms.
TLDR here - all revenue flows to foundation (no equity), all reported revenue checked by a 3rd party (report attached)
TLDR here - all revenue flows to foundation (no equity), all reported revenue checked by a 3rd party (report attached)
Grass DataCo Ltd., a wholly owned subsidiary of Grass Foundation, provides data infrastructure to leading AI labs training frontier models.
We're publishing an independent attestation of its revenue through Q2 2026, conducted by Regen Financial.
Full report: grass.io/resources/gras…
86
16
177
32,407