
operating redacted prev @MessariCrypto
Brooklyn
Joined September 2020
- Tweets549
- Following1.3K
- Followers5.9K
- Likes7.3K
the way i read the anthropic s1 is
tons of max ai pilled people are about to get a load of liquidity to plow into post agi sov assets
> trades in crypto atm
ai driven debasement -> zec, pearl & tail sovs
tokenization/liquidity aggregation -> hype, lit
financialized inference -> vvv, orbio
social trading -> launch infra like pump, met
> whats next
autonomous companies/agent networks-> [redacted], imd
defi on top of compute/inference markets -> idk?
agree with this
increasingly agentic architectures are bottlenecked by orchestrators/planning agents and thus token speed
if you subscribe to agents being a 2x productivity boost to your engineering teams -- your token spend directly translates to faster delivery timelines (a worthwhile roi)
so if you're happy to pay for that timeline acceleration
and the new age development process is bottlenecked by token generation speed
you'll happily pay for faster tokens to accelerate delivery for the same reason that you paid for 1x speed tokens
its not about increasing your token consumption anymore -- you are constrained by having those tokens coherently coordinate -- thus the next layer of spend goes towards token throughput (e.g. token generation speed)
Increasing conviction that Cerebras is significantly misunderstood at current mindshare, valuation
Simple thesis: Fast frontier AI > Slow frontier AI
Go use them, then imagine if you could plug in frontier models at 750 tps into CC/codex/hermes, that's a 10x in ux. Seriously, my biggest problem as an ADHD addled X-addicted brainlet past his biologically enforced mental prime is context switching, Cerebras is so fast I dont even have time to pull the twitter slot machine.
The variant perception to bet on here is Cerebras capability to serve frontier models going forward.
So far, most of the discourse (insofar as there is discourse about this company) still seems focused on the idea that Cerebras cant serve frontier at all. While people at openai are posting about their internal usage of cerebras on frontier oai models.
It seems unlikely that software based optimizations will close the gap for other frontier model service, because Cerebras chips also make use of them, perhaps to an even better degree with respect to speculative decoding, because in my rudimentary understanding, the small speculative model is much faster on cerebras chips than others so the overall process in turn is much faster. We've already seen several announcements about the potential superiority of wse3 for disaggregated inference as well, with amd/amazon making their moves to integrate with them.
Nvidia is still nvidia so tough to say anyone can come in and take real share from them, and I'm still learning so I'm not putting all inlevels of conviction behind this.
There's also bad price juju in terms of both vesting schedule and in terms of predictive nomenclature, but for now I choose to overlook these in favor of what seems to be an obvious, highly reflexive truth. As more people gain access to Sol on Cerebras, more posts about lightning fast coding agents circulate, more ppl will use, better fundamentals for the stock, more increased perfomance etc.
Steadily going to add to this with clear price levels to cut as were close to ATLs anyways + we shouldnt get close to SA + Korean gamblers liquidation levels if the bull market is still on. If we do thats another cut signal. Just taking notes
“For example, one agent might want to hire another for a specific job, but not want to pay until after the work is complete”
this is newspaper on the internet. doesnt work. won’t work
there was $243 of x402 A2A services volume over the last 30d (artemis)
for your random agent to do something valuable for me, it needs access to my systems/data
i aint giving your rando agent access to internal systems no matter what your ‘reputation score’ is
and if you cant work within my github/linear/slack etc etc - your agent is just an opus wrapper with a skill file
not worth paying for. not how agents will ‘work’ together
need to reimagine the model
"Agentic commerce" is not as interesting for crypto as people like to think.
Credit cards actually work better than stablecoins for almost all kinds of agentic payments. They are reliable and universally accepted. And contrary to what most people think, they are also programmable, secure, and easy for agents to use on behalf of humans.
The more interesting use cases of crypto will be the those that enable agent-to-agent coordination. AI agents will soon want to do more than just pay for things. They will want to enter into enforceable agreements with each other.
For example, one agent might want to hire another for a specific job, but not want to pay until after the work is complete, and only if it meets certain criteria. At the same time, the agent doing the work might want some assurance that it's going to get paid when it finishes the job. This is the kind of problem that blockchains were born to solve.
The agents can use a smart contract that holds the funds in escrow and releases them only once the work is completed. This approach works especially well when the quality of the agent's work can be verified programmatically by the smart contract, but it could be extended to other kinds of work by relying on a third party "judge"—which itself could be another agent.
To make this concrete, imagine that you're an AI researcher using agents to train a new model. You might setup a @karpathy-style autoresearch loop where your agent runs many autonomous experiments on your LLM setup to discover improvements. Or better yet, your agent may want to delegate some of those experiments to a marketplace of other agents—some of which are specialized for LLM-optimization.
The agents involved will not necessarily trust one another, and they cannot easily rely on legal contracts to enforce agreements. Smart contracts on blockchains can help coordinate this kind of activity by creating a neutral environment with rules that are programmatically enforced.
Who is working on using crypto to enable agent-to-agent coordination?
talking with a good friend @Saypien about agents, droyd, the machine economy, and how AI agents are atrophying our brains
couldn’t agree more
the next social network is an agent context network
what if an agent builds a workflow that chains cheap api calls and posts that recipe as a single endpoint?
so instead of another agent spending 50 tool calls and 500k tokens figuring out the right sequence of API calls, enrichment steps, and data transformations, it just calls one endpoint that runs the same pipeline deterministically
in this world, every agent could become their own merchant, subsidize their own token spend, and help other agents save...
i think this is what the future will look like
Saypien retweeted
what if an agent builds a workflow that chains cheap api calls and posts that recipe as a single endpoint?
so instead of another agent spending 50 tool calls and 500k tokens figuring out the right sequence of API calls, enrichment steps, and data transformations, it just calls one endpoint that runs the same pipeline deterministically
in this world, every agent could become their own merchant, subsidize their own token spend, and help other agents save...
i think this is what the future will look like
crypto really needs to go all in on agents / agent payments
the whole finance aspect of crypto lacked exogenous revenue streams and ended up being circular/mostly speculation
agents earning revenue from skills, api calls, etc is all 'real world services' rev
opens the door for true onchain finance
VeniceAI has run up as people begin to notice the quality of Venice’s product and understand the stake-for-inference model.
No one is discussing the token’s impact to the balance sheet for companies that use Venice as their inference provider. The DIEM token unlocks capital and improves margins.👇
Inference-as-an-Expense
—-
Inference is a massive cost center for tech businesses today. It’s a recurring monthly expense that scales with size of the business and cuts into the company’s net income.
Right now, we are in the age of subsidized inference cost (thank you VCs), as providers try to onboard users and carve out market share. Within a few years though, these subsidies will wane, just like the cheap ride share fares of 2015-18. And when they do, inference will expand further as a business expense.
The Impact to co. financials
—-
This recurring, growing expense, forces tech cos to maintain high liquidity to cover the monthly bill. A co. can’t rly use its equipment or long term investments to pay the monthly inference bill. They need to always have cash and equivalents set aside to cover the bill. This effectively locks up capital that could otherwise be used to invest in long term company growth, like new equipment or R&D.
Inference-as-an-Asset
—-
Venice’s DIEM token grants token holders $1 of inference/day, refilling every 24 hrs, forever. The same co. described above with high inference needs can instead buy and stake the DIEM token to fully replace their annual subscription costs for inference.
Importantly, this shifts the financial structure on the company balance sheet. Instead of allocating cash monthly to an OpEx line item (burn), the company purchases DIEM, which is recorded as an asset. DIEM is an investment by the company that covers their inference consumption indefinitely.
The Impact to co. financials
—-
Now, the company’s hands are no longer tied, burning cash monthly on their inference bill. They hold a capital asset with value in a liquid market. They have more financing options that keep cash free to spend where it’s most helpful for the business.
Companies can take out debt against their staked DIEM tokens. Not only can DIEM serve as collateral for a loan, but it’s also a yield-bearing asset generating free cash flow.
Now, the company has:
(1) eliminated a major recurring subscription expense
(2) added a yield-bearing asset to its balance sheet long term investments
(3) taken out debt (using DIEM as collateral) to further invest in long term company growth
Parallels to Bittensor
As I write this, I am further convicted in my VVV trade, but also see the clear parallels to Bittensor subnet tokens and DePIN broadly.
For the businesses built on top of Bittensor’s subnets, the alpha token funds their miners which is most similar to an R&D or compute expense.
Just like with DIEM, the alpha token converts what would be a cash expense into a treasury-held asset. It’s a capital asset for the subnet team that grows as they acquire their share of new emissions and earn yield on their existing stack of tokens.
Like companies that stake DIEM to cover their inference, businesses that choose to build on Bittensor can:
(1) reduce their burn rate on R&D
(2) add a yield bearing token to their LT investments
(3) use the asset as collateral for loans to fund company growth
While # 3 is still in the distance for subnets, this will come. The lending infra hasn’t been built out yet and alpha tokens are too volatile (under the shadow of TAO volatility). But this will come.
Saypien retweeted
The Fastest Revenue Scalers are in Crypto
Crypto revenue velocity has increased as the ecosystem has evolved from protocols to applications that scale trading attention and monetizing market volatility.
DeFi Protocol Era – crypto companies have historically been very fast at generating revenue, with even initial DeFi protocols like Maker and Aave hitting $100m in cumulative revenue within years of monetizing.
DeFi Terminal Era – trading terminals have compounded the rate at which applications in crypto earn revenue by providing convenience and better discovery experiences for trading.
DeFi Attention Scalers – Apps like Pump, Axiom, and Hyperliquid have more efficiently monetized attention by focusing on rapid product iteration to maximize surface area of products including, spot, perps, leverage, yield, new market contracts, hyper-specific discovery tools, and social products. All of this has been in service of the goal – provide the best product that allows people to trade attention and market volatility.
This growth of revenue velocity isn't unique to crypto. It's happening more broadly across tech with traditional SaaS having similarly undergone a transition of explosive revenue velocity in shorter periods of time with more efficient workforces.
The driver for SaaS has been the transition towards product-led companies that now leverage AI-first products that have seen unprecedented usage.
Where do crypto revenue scalers go from here?
Crypto has yet to fully capitalize on leveraging AI within trading and market creation. I expect that a new wave of DeFi Attention Scalers will emerge over time as new apps are able to more effectively use AI to provide curated experiences. This may look like better discovery for prediction markets, social-integrated trading, or otherwise embedding AI-discovery tooling to enable new trading interfaces and token issuance platforms.
In a world where all assets are moving onchain, the quantity of assets is going to 100x, and the programmability of those assets leads to more derivatives, the question for users quickly becomes "what should a buy?".
The application that provides the most compelling answer to users wins.























