I’m thrilled to share that I’ve been promoted to General Counsel & Investment Partner @variantfund! As GC I’ll continue the work I’ve been doing alongside @jchervinsky to help our industry define legal & regulatory strategy, & going deep (code-level) w/ founders on their product ideas. I'm also excited to continue working closely w/ founders as a member of the investment team. If you’re working on something new, I’d love to hear from you!

New piece: Why (good) AI needs crypto. TLDR: Closed AI is anti-competitive. Open source AI hits a “resource problem” (huge compute/data costs). Crypto fixes this by incentivizing resource contributions via ownership—unlocking bigger models & faster innovation. 🔗👇 variant.fund/articles/why-go…

Good crypto x AI projects leverage the features of crypto to overcome limitations faced by AI. Here are three examples of challenges of AI: 1. Resources. To be competitive as an AI company you need to aggregate significant resources (compute and data) and talent. 2. Composability. Agents cannot reliably chain actions together. 3. Native Payments. Agents, while nearly infinite in their ability to act on a web made up of natural language, remain restricted by the fact that payment rails do not exist on internet rails. These challenges can be solved through separate features of crypto: economic ownership, verifiability, and self-custody. Economic Ownership Crypto is amazing at using ownership to aggregate resources and talent. This is really just another way of saying crypto solves a coordination problem. Aggregate Resources: Use ownership to incentivize compute and data providers to contribute their resources. These resources can be used to build a shared data layer between siloed applications, bootstrap the high upfront costs of training, or attract new sources of supply for inference/training. Examples: @plasticlabs (shared data layer), @PluralisHQ (training), @hyperbolic_labs (inference/training). Aggregate Talent: Use ownership to aggregate talented participants to exert labor/skillset to a shared layer. There’s two reasons this matters. First, diverse talent at a shared layer leads to high variance — and when paired with composability (which is enabled by another feature of crypto, verifiability) can produce interesting emergent systems, like agent swarms. Second, this human labor can be used to refine data used by AI, like for data labeling or RLHF of taste preferences. Examples: @itsalltruffles (agent swarms), @PlaySapien (data labeling), @bottoproject (model taste via RLHF). Verifiability Trust is inefficient: it requires further mechanisms to ensure the trusted party does not abuse that trust, like legal contracting/putting a human in the loop. With cryptographic verifiability, frictionless composability becomes possible. This will be particularly important in the context of agents who need to act autonomously to carry out various tasks, many of which are undetermined at the time the agent is spun up. Composability is a key ingredient to make agent swarms possible/effective. Examples: @eigencloud (verifiable agents) Self-Custody The primary unlock of self-custody with AI is currently around agents. Giving an agent a wallet — integrating payment rails into internet rails — provides it with financial independence and therefore an unprecedented level of autonomy. This enables great UX, allowing a human to delegate to the agent (within set bounds) a complicated series of transactions without needing to put a human in the loop to pay each step of the way. The best examples of agents that utilize this feature are DeFi agents and services agents. Agents with something closer to true autonomy are also just entertaining (whether for the novelty or intrinsically). Examples: @AIWayfinder (DeFi agents), @autonolas (services agents), @truth_terminal (entertainment agents).

I am unaware of an area of law with more bang for its buck for crypto apps than a website’s terms of service. I know terms of service are not the most exciting thing in the world, but if done right, they can make a huge difference in legal liability exposure. The key issue apps run into in this area of law is whether the terms are enforceable on the app’s users. So, I looked into tips for increasing the likelihood of enforceability. Here’s a thread with the TLDR.

One way to look at NFTs is that they are valuable because they allow digital property to reflect features we associate with property in the physical world. Think about owning a house. When you buy the house, you get a deed, which provides provenance via a chain of title. As the owner, you unilaterally possess the property and have the right to exclude others from using it. And if you improve the house, those improvements run with the property. Normal digital assets do not have these features, but NFTs can embody them. NFTs have so far only reached product-market fit with art, which generally only utilizes the first two features. Take Asymmetrical Liberation, a piece from Botto. It certainly utilizes provenance to prove it was the first piece generated in Botto’s Genesis Period, and its holder has unilateral control over it. But there’s no way its holder can improve the art (nor would that make sense). The closest NFTs have leaned into this improvement feature is probably with video game NFTs, like when a player’s action in the game levels up a character. This feature is far too underexplored with NFTs. I think one really interesting use case for NFTs that can impart all three of these features of property in the physical world is agents. The key insight is that agents have “memories” that allow them to record their experiences performing tasks for their principals. These experiences are unique, which makes them non-fungible. An NFT could represent an agent, and holding this NFT could provide exclusive read/write access to an associated memory slot in a trustless memory layer (functioning effectively as a “license” to these memories). This memory layer is exactly what companies like Plastic Labs are working on. Agents as NFTs would embody the three features of property in the physical world: Possession. The holder of the NFT will have the exclusive right to access the agent’s memories. Improvements. The agent’s memory will update every time the NFT holder uses the agent, gaining holder-specific experience. The agent will also remember the refinements of the holder. Provenance. Because the agent can only be used by its NFT holder, a record of past holders will indicate its experience, functioning like a “resume.” Leaning into these features of property in the physical world will allow for a highly specialized and differentiated agent that isn’t possible today. For example, imagine J. R. R. Tolkien held a certain NFT agent and used it while writing The Hobbit. Let’s say you want to write your own fantasy novel and are seeking an agent to help you. The fact that this agent was provably owned by Tolkien, and only Tolkien, indicates it has experiences that will be helpful for drafting a fantasy novel. If you acquire it, the agent would have added value because only you will be able to access it; the agent is a differentiator for you. While the majority of agents will probably make more sense as collective agents offered to the public via an API like any SaaS product, agents as NFTs enable something more akin to a racehorse. Racehorses are more valuable when they provably come from certain trainers, their training stays with the horse from owner to owner, and they can only be possessed by one rider at a time. Agents as NFTs unlocks a net-new design space in software I’m excited to see play out.

About 6 months ago I read a summary of a 2nd Circuit case where the court held that Binance was subject to the U.S. securities laws, in part, because it used AWS instances in California. My immediate reaction was that I misread the summary. I reread it; nope, I read it correctly. I then assumed the author made a mistake, so I read the opinion myself. Nope again, server locations were core to the court's reasoning. This sent me down a rabbit hole. How could the location of a server be so important? So I read every crypto x securities x jurisdictional scope case I could find to try and understand. I became obsessed. Today I'm excited to publish with @jchervinsky the fruits of this obsession. We've distilled everything we learned by reviewing tons of case law and regulatory enforcement actions into a Practical Guide to Geofencing for crypto companies. I hope you enjoy. Check it out here: variant.fund/articles/practi…

Agent devs need to grapple with the general rule crypto devs have had to deal with for years now: Custody user funds = increased liability risk Why? Control is the law’s lodestar for finding liability Two issues immediately come to mind

New article unpacks the unique liability considerations for agent developers. TLDR: agents introduce complexity but there are some common sense measures developers can take to reduce their liability exposure. Graphic below summarizing that. Check out the piece here: variant.fund/articles/liabil…

Today, we’re excited to announce our pre-seed investment in @plasticlabs. The rise of LLM apps is causing unprecedented demand for personalization in software. These apps rely on natural language, which changes depending on who we’re speaking to. Think about how you would explain a math concept to your grandparent compared to your parent or your child. You instinctively adjust your explanation based on the listener. LLM apps must similarly understand who they’re talking to in order to deliver more effective, tailored experiences. Whether it’s a therapist agent, a legal assistant, or a shopping companion, these apps must develop an understanding of their users to be truly useful. Despite how essential personalization is, there’s currently no out-of-the-box solution for it for LLM apps. Developers are forced to build ad hoc systems that store user data — typically as session logs — and then retrieve that data as needed. This leads to every LLM app redundantly solving the same core problem of building infrastructure to manage user state. Worse yet, many of these solutions, such as storing user interactions in a vector database and running RAG over them, aren’t actually capturing the user — they’re capturing the user’s interactions. While helpful for recalling past conversations, these approaches offer little meaningful insight into who the user actually is: their interests, communication preferences, tone sensitivities, and more. Plastic Labs solves this. The team has built Honcho, a plug-and-play platform that lets developers personalize any LLM app effortlessly. Instead of reinventing user modeling from scratch, developers can integrate Honcho to instantly gain access to rich, persistent user representations. These representations are far more nuanced than anything possible with traditional methods, thanks to techniques borrowed from cognitive science. They are also fully queryable via natural language, allowing LLMs to flexibly adapt behavior based on this underlying user representation. By abstracting away the complexity of user state management, Honcho enables a new level of hyper-personalized experiences for LLM apps. But the implications go even further. Honcho’s ability to generate rich, abstract representations of users also unlocks the long-elusive promise of a shared user data layer. Shared user data layers have historically failed for two main reasons, both of which Honcho addresses: 1. Lack of Interoperability: Traditional user data is context-specific and not portable across apps. For instance, X might model you based on who you follow, but that data isn’t useful to who you are professionally connected to on LinkedIn. Honcho, by contrast, captures a higher-order and universally relevant representation of the user that works seamlessly with any app that uses an LLM. For example, if a tutoring app discovers that you learn best with analogies, your therapist agent could use that same insight to communicate more effectively, even though these apps have completely different contexts. 2. Lack of Immediate Utility: Previous attempts at shared data layers struggled to bootstrap because they didn’t provide value to early adopter apps, which are where useful user data is generated. Honcho solves this by being directly useful to individual apps, even in “single-player mode.” Apps adopt Honcho to manage user state — a first-order problem — and will eventually be able to contribute to a shared layer. Once enough apps contribute to the layer, the network effect kicks in: newer apps will integrate not just for personalization, but also to tap into the collective intelligence of shared user representations, solving cold-start problems and enhancing UX from day one. Honcho already has hundreds of apps on its closed beta waitlist — from sobriety coaches and educational companions to reading assistants and e-commerce tools. The team’s strategy is to focus first on solving the core challenge of user state management for apps (the first-order problem), then roll out the shared data layer for interoperability (the second-order problem) for those that opt-in. This layer will use crypto to align incentives: early apps that contribute to the system will receive ownership in the layer and, therefore, upside in it. Crypto also makes the system trustless, addressing fears that a central authority could extract value or build competing products. We believe Plastic Labs is the team to tackle the challenge of user representation in LLM-powered software. @vintrotweets (CEO), @courtlandleer (COO), and (CTO) discovered the problem by experiencing it directly while building personalized experiences into their chatbot tutoring app, Bloom. Their students weren’t learning effectively because the app did not fundamentally understand who the students were and how they learned. Honcho was born from that insight, and it’s solving a pain point that every LLM app developer will face.

Some thoughts on agents and "memory."

You know our founding thesis. Invest in an internet that turns users into owners. The way I've typically framed ownership is through provable digital scarcity. While this framing is true, @metaproph3t has opened my eyes to the fact that it's incomplete. Because behind every project is still people. And if people have no accountability, ownership rings hollow. MetaDAO merges these concepts to represent the best shot I've seen for ownership to mean something online.

This post nails it - founders need discretion for @MetaDAOProject to be competitive. This latest round of fundraising is a microcosm of the existential question as to the role of governance in the project. Thread on my thoughts, coming at it as a former founder/lawyer 🧵👇

In my latest post on my blog Proofs and Protocols, I teamed up with my colleague, Variant Investment Partner Cooper Kunz (@cooper_kunz), to dissect the Computer Fraud and Abuse Act (CFAA) and browser extensions (e.g., Chrome Extensions). Here’s a thread with the TLDR 👇

I’m thrilled to announce our seed investment in @PluralisHQ. Pluralis is developing true open-source AI through decentralized training. Variant’s founding thesis is that crypto enables an internet where users become owners—solving coordination problems by aligning incentives through shared ownership. One coordination problem crypto excels at solving is the “resource problem,” which occurs when software development requires costly external resources beyond an individual’s means. Our seed investment in Pluralis Research is premised on using crypto ownership to solve the resource problem of foundation AI models. Let’s unpack that. The Resource Problem of Foundation AI Models and the Cost of the Corporate Form It costs tens of millions of dollars in compute and data to pretrain a foundation model. Foundation models therefore suffer from a severe resource problem that to date has only been solved through the “corporate form”: closing off the software into a corporate entity and raising a massive amount of capital to get over the resource hump. As a result, only a few companies have been able to raise the capital necessary to develop foundation models—all built behind closed doors. Even Meta and DeepSeek, which have released their foundation model weights publicly, build their models privately with their own resources and determine when to make them available to the world. Using the corporate form to solve the resource problem comes with significant costs. Looked at normatively, if only a few entities can build foundation models, they will dictate access to information and opportunity, setting us up for a dystopian hellscape of a magnitude Orwell couldn’t have dreamed of. But there is also a major opportunity cost, as our models will forever remain suboptimal because of constraints on talent and resource supply. 1. Talent Constraints Today, to build foundation models, you must work inside one of the major corporations whose research labs steward their development and training. These corporations have limited seats and highly selective hiring processes, indexing heavily on candidates with fancy pedigrees. This necessarily excludes a whole suite of high-quality model designers with nontraditional backgrounds. Additionally, many high-quality model designers do not want to work inside one of these organizations for personal, ethical, or geographic reasons. When the talent pool is global, the corporate form excludes a significant chunk of talent. This forecloses a diversity of ideas and new thinking for how to build foundation models. These corporations also constrain the collaboration of talent across and outside their walled gardens. Model designers cannot iterate on each other’s work fluidly, which contradicts how machine learning development has historically occurred. This hampers innovation and limits our ability to reach the next frontier of model design. 2. Resource Supply Constraints Under the current system, AI compute and data resources are fragmented and trapped inside the walled gardens of the few entities that can afford them. Each major model provider hoards these resources, creating redundant silos that limit collective progress. Because each entity is closed, there is also no way for nontraditional resource providers to contribute things like consumer GPUs and data on individuals’ devices. This locks out a significant chunk of valuable resources that are otherwise sitting idle. Current research suggests that, all things being equal, more compute and data in pretraining means better models. By failing to tap into the full potential of supply, our models’ performance will remain artificially stunted. Pluralis Uses Crypto Ownership to Solve the Resource Problem If we used crypto ownership instead of the corporate form to solve the resource problem of foundation models, we would unchain these constraints on talent and resource supply and build better models. This is exactly what Pluralis does. Pluralis is building a decentralized training protocol where model designers can come to the protocol with their model ideas while compute and data providers can contribute the resources necessary to train models. These “protocol models” are open, developed in public, and owned by a network of resource contributors. The project uses the two core features of crypto ownership to get over the resource problem of foundation models. First, ownership in the form of a token is provided directly to speculative compute providers who contribute compute to train a model. This token represents upside in the model in the form of a claim on the future inference revenue it will generate. Second, ownership is made meaningful but the software remains open because the value of the software—the weights that represent training—is moved onto a distributed network. To do this, the project exploits a key property of neural networks: their computations are local, which means their weights can be split up across the nodes of the network through model parallelism. This second feature represents a really thorny problem because it requires solving challenges at the intersection of distributed systems and AI. This intersection is largely unexplored because big tech incumbents have not needed to explore it; as centralized entities, they can assume homogenous and performant GPUs that are largely centrally hosted with high-speed interconnects. This is at odds with the permissionless and globally distributed system that Pluralis requires, which involves heterogeneous GPUs with highly variable speed and memory, interconnects of varying bandwidth and reliability, and large distances between nodes. The Pluralis team is the one to tackle this novel research problem. Led by founder @alexanderlong, the team currently consists of five machine learning PhDs who all left AI research at Amazon to find better ways to build models. They have authored foundational machine learning papers and are world experts in relevant research areas for distributing training, such as implicit neural representations (highly relevant for compression). They are approaching the problem from first principles rather than assuming a centralized stack, building on recent research that suggests training over lower-bandwidth networks is feasible. We were deeply impressed by the progress the team has already made in its novel research around distributed model parallelism. We believe the models built on Pluralis will be better than our current models built under the corporate form because they will not be subject to existing talent and resource supply constraints. By drawing a line of ownership directly between the model designers who need resources and the providers who have those resources, Pluralis will serve as a talent marketplace for compute and data providers. Tapping into the upside of model designers previously trapped inside labs will motivate new parties to serve as resource providers, just as the prospect of Bitcoin ownership forged a new mining industry. This will aggregate resource supply levels that are simply not available under walled gardens, leading to bigger and better models. And because the models will be publicly developed, model designers will be able to iterate on each other’s work, leading to more innovation in model development. We could not be more excited to invest in Alexander and the Pluralis team as they use crypto’s superpower of ownership to solve one of the most important and challenging resource problems in software: making better AI that is by the people, for the people.

1/ What are agents good at? We debated internally and came up with at least four things: (1) meeting humans where they are; (2) doing work for nudging humans; (3) aggregating and synthesizing information and (4) being entertaining.

1/ Computing platform shifts often evolve in couplets or triplets—new hardware, applications, & distribution innovations propelling one another forward. Think: - PC, Internet, Web - Mobile, Social, Cloud We believe a new pair is emerging: crypto & AI. Let’s unpack this. 🧵👇

If you're attending @SolanaConf, Variant is hosting co-working with Phantom on May 21 at our office Ship code and meet other amazing builders Space is very tight, sign-up here: lu.ma/xqouspi6

It was a teeny weeny footnote in our 40 page analysis on 1960 with @jchervinsky and @amandatums, but I was always struck by the DOJ's brief in a case called Faiella, where they literally describe "transfer" (the relevant term in "money transmitting") in 1960 as "effect[ing] a change in the possession or control over . . . funds". Yet, in cases like Samourai and Tornado, they argue that reading control in the statute is borderline preposterous. 🤔 Full paper here: edit.financialcrimelitigator…

One way to think about the case for crypto x AI is it’s a bet on horizontal v vertical scaling. Big tech has mastered vertical - hoard compute, data ,and buy out competitors and vertically integrate them - form a monopoly. Crypto horizontally scales - high variance in contributions, massive consumer compute, data, etc. Bitcoin as the largest compute network shows horizontal can outcompete vertical. But the contributions are relatively deterministic (does this nonce hash). When things are deterministic it’s easier to scale infinitely horizontally: determinism —> easy verifiability —> permissionless —> anyone can contribute. That’s how you get anyone with a GPU to contribute compute. The hard part about AI is a lot of useful work in the real world that needs AI is not so cleanly verifiable. So we have two options to make this scale: 1. Focus on parts of AI systems that are more easily verifiable with cryptography like a zk proof showing you contributed compute. 2. Come up with new ways to provide for fuzzy verifiability through crypto economics, like using AI itself as a neutral fuzzy arbiter. There’s no other way to do it.

There are three pillars to crypto x AI: aggregation, verification, and self-custody. Loved diving in to this with @jeffwilser - excerpt below and full podcast for the brave here: open.spotify.com/episode/6bN…

We’re excited to co-lead @TakenosApp seed round with @lattice_fund, backing a team that’s redefining access to digital dollars across LATAM. Takenos enables users to receive money from abroad, save in USD and spend globally through a virtual dollar card - all powered by stablecoins. Takenos has reached a level of PMF with real world adoption that has long been the holy grail of crypto companies. They have consistently been a top 5 app in Bolivia in the App Store (yes, all apps, not just crypto apps) - at times, even more popular than ChatGPT. Unlike many LATAM fintechs focused on the “big three” markets of Brazil, Mexico, and Argentina, Takenos is winning across the long tail - countries like Bolivia and Peru that have been traditionally neglected by both banking and fintechs alike. The founders, Lucas, Francisco, Joaquín, Simón, are builders in the truest sense: hands-on, resourceful and willing to relocate their entire team to each new market they enter. We’re proud to back the Takenos team as they build the future of onchain finance in LATAM.

Really important development - DOJ memo calls for an end to reg by crim enf. Was perplexed by fn 2 which says prong (C) of the criminal MT statute is "outside the scope of this policy" - that prong is and remains deeply concerning to me. But overall, very positive development.

I will continue to beat this drum - there is no lower hanging fruit for a crypto company to get right than strong/enforceable TOS.

Jake and Salah are the dream team. We are consistently seeing crypto companies that are ahead of the curve on policy win. You should talk to them.

That's it. That's the tweet.

Agents. As. NFTs. blog.variant.fund/agents-as-…

When all you have is a hammer, everything looks like a nail. On how we can use new primitives to solve squishy problems instead of putting the burden on DAOs.

The @plasticlabs team is shipping with such high velocity right now it’s honestly hard to keep up. If you want SOTA memory in your app - outperforming even Claude 4 Sonnet - you need to check out @honchodotdev.

Agree - memory is the moat when the model is the commodity. Apps have a small window to offer personalization to users before ChatGPT builds too much stickiness. So if you’re an app you should be talking to @plasticlabs and using @honchodotdev STAT. *~*~end of message~*~*

Agents + TEE can be used as neutral arbiters of truth, particularly useful when the thing being arbitrated has fuzzy inputs. Could be unlock for DAO 2.0s - DAO sends agent what it wants resolved, it gets final say - grants, dispute resolution, etc. Who's building this?

So excited for this event tonight. No fluff - going to get into the weeds of opportunities/challenges around agents. If you're in NYC, stop by.

This is awesome. I used to use Etherscan as an associate at a law firm and it was bafflingly complicated. I would have to write custom scripts to answer the simplest questions. I literally cannot imagine how much time I would have saved by using something like this.

Discussion forum on crypto agents went well today . . . .

None of the above is legal advice (as always), these tips are not a silver bullet to making an enforceable online contract, and this is an area of law in flux. You should discuss your situation with your legal counsel. Learn more in my piece: blog.variant.fund/enforceabi….

Looking for projects that are leveraging sybil resistance of World into a protocol.

If only there was an article that unpacked liability considerations for agent developers . . . . variant.fund/articles/liabil…

This assumes the future is monolithic (ChatGPT eats the world). But I don't think that's right Apps don’t want to be locked into a single model/have their data siphoned to big tech But to compete, apps need an identity layer to understand users across apps That’s what @plasticlabs is building

This is an insane accomplishment by the @PluralisHQ team. ANYONE can contribute to training a model ANYWHERE in the world with a consumer grade GPU - fully permissionlessly. Fundamentally changes what future we choose to live in.

Money transmission remains the most existential issue. If software devs can be CRIMINALLY liable even with NO CONTROL over funds there is no limiting principle. We're fortunate for @fund_defi and these other great orgs taking on this fight. We need to get this right.

One area I’m particularly excited about is using TEEs as a way to remove agent dev control @turnkeyhq does amazing work on this front - you should talk to them @flashbots_x is also doing amazing work on TEEs

I'm going to be at the Port next week - if you want to meet up or chat let me know my DMs are open.

Most people think the role of stablecoins in U.S. consumer payments will only ever be a backend settlement tool to make credit cards more efficient. But I think they can replace cards altogether. They can do that by following the same two-step path Visa used to achieve mainstream adoption: (1) lean into intrinsic advantages to solve real pain points for specific users, and (2) build or participate in an open network that aggregates fragmented usage into a scalable, interoperable system. Step One: Start with niche use cases by leaning into intrinsic advantages. Credit cards overcame the bootstrapping problem by focusing first on intrinsic features — convenience, incentives, and increased sales — that didn’t require a fully built network. Stablecoins can do the same by finding early traction in two niches: (1) Comparative convenience: For consumers in jurisdictions with restricted USD access (e.g., LATAM), stablecoins offer uniquely convenient access to U.S. merchants, enabling previously impossible sales. (2) Incentive-first use cases: Whitelabeled stablecoins can power rewards programs, with merchants offering discounts and perks funded by yield on the float. Consumers are often willing to tolerate onboarding friction for products they love if the rewards are strong and the value is likely to be reused (e.g., Starbucks Rewards, Taylor Swift fans, or Poshmark users). Step Two: Aggregate fiefdoms into a full-blown network. As in the early credit card era, leaning into niche use cases will lead to fragmentation — varied chains, issuers, user experiences, and standards around consumer protection. Over time, these fiefdoms will need to be connected through a neutral, interoperable network, a “Visa for stablecoins.” Stablecoins won’t dethrone cards by competing head-on. They’ll get there by serving edge cases first, then stringing those edges together. This will aggregate the supply and demand necessary to solve the bootstrapping problem of a new payment method. From a new consumer’s standpoint, joining the stablecoin world will eventually offer enough value that the one-time onboarding pain becomes worth it. At that point, stablecoins won’t be seen as an alternative to credit cards. They’ll be seen as the inevitable successor. Read the full piece here for my detailed argument, including: - The similarities between Visa’s early history and our present moment - Why merchants would choose whitelisted stablecoins over traditional rewards programs - How customers weigh convenience and incentives

Thank you to @alexanderlong, @zjasper, @bidhan, @shivani_3000, @_jamico, @entropybender for thoughtful feedback and inspiration.

13/ Today, we’re excited to announce that we’ve led @hyperbolic_labs Series A round: blog.variant.fund/investing-… We believe in their mission to make AI more accessible, verifiable, & open.

AI AND CRYPTO AT VARIANT GREAT FOUNDERS GREAT PROJECTS ME AND @jphackworth42 JOIN US

Someone is impersonating me under the handle "@barabander" on TG. That is not me. Please report this account and do not interact with them.

While it sounds nice, I think this will not be able to compete with stablecoins long term. SWIFT does not solve the issue that we are dealing with different trust layers. The hourglass market remains.

What am I missing? What are other limitations of AI we should be using crypto’s strengths to solve? @alexanderlong @iamgingertrash @zjasper @RowanRK6 @hudsonsims @vintrotweets @sreeramkannan @annakaz @bidhan @shivani_3000 @mo_baioumy @vincentweisser @ilblackdragon @gregosuri @Shaughnessy119 @0xPrismatic @Ronangmi @jonathankingvc @yb_effect

TELL US HOW WE'RE WRONG

YB doing the lord's work. Thanks for this - excellent piece.

But presenting your users with a wall of legalese is bad UI, so over time, apps have increasingly hyperlinked to the terms (instead of displaying the terms on the page itself) and included an action item to manifest assent, even if this is generally seen as riskier than a pure clickwrap. This is commonly referred to as a “sign-in wrap” or “hybrid wrap,” but I find that name confusing, so I call it a “hyperwrap.”

If you’re a Web3 company building a browser extension and find this interesting, please reach out; would love to hear what you’re working on

We're looking for customizable agent launchers that abstract away putting those agents on rails. Platform's target customer is someone who wants to build really interesting agents that are differentiated and specialized for particular tasks. DM me if you're building this.

Thank you to @plasticlabs, @memitations, and @delta_alpha_ohm for their thoughtful feedback on this piece.

Generally speaking, the “gold standard” for terms of service is a “clickwrap” or “scrollwrap” (I’m treating these as similar things, but different people call these different things). This is when the app presents the full terms, and users click a box saying they’ve read and agree to them. As one court put it: “To ensure that an online agreement passes muster, clickwrap is the safest choice.”

So excited that @JackLongarzo is joining us. He is the real deal. A true builder and has one of the sharpest minds I've encountered on GTM/product to help advise founders. An amazing addition to the team.

Link to the article here: blog.variant.fund/agents-as-…

Courts take an extraordinarily detailed and design-centric approach when assessing the enforceability of hyperwraps, focusing on whether they provide adequate notice to the user about the terms and whether the user’s action manifests assent. Based on my review of important cases, here are 5 tips to increase the likelihood that the hyperwrap is enforceable.

As you'll hear, I'm the most bearish on the team for the "everything exchange." Tell me why I'm wrong.

Insane opportunity to join a cracked team building on HL.

I am a proud to be a cactus.

Fourth, present your hyperwrap at critical stages like sign-in, sign-up, and time of purchase. This helps provide notice and reinforces the context of a continuing relationship, which courts look for in enforcing terms that contemplate as much.

If you're vibe coding, building agents or any other app and want to manage user state/get personalization with an API that lets you query that state WITH NATURAL LANGUAGE you need to check out @honchodotdev

First, present the hyperwrap on a clean, uncluttered page. The hyperwrap should be the star of the show when it’s presented, not buried.

2/ AI requires the massive coordination of GPUs and crypto uses incentives to herd resources. AI is probabilistic and crypto is deterministic (h/t @balajis). We believe crypto can solve two of the most pressing problems for AI: cost and trust (and specifically, cost of trust).

Caleb is an absolute killer and we could not be more excited for him to join the team.

Greatly looking forward to joining @jchervinsky and the @exa_bits team for this AMA today at 12 PM ET! Going to cover some of the unique legal wrinkles at the intersection of crypto x AI. Join us soon!

A number of our recent pre-seed and seed portfolio companies are ACTIVELY HIRING. These companies are building the future of crypto. If you want to get in on the ground floor, fill out this short form and I will try and connect you with a fit. 👇

Really looking forward to chatting all things agents - sign up below to attend 👇

We actively looking for projects in crypto x AI that lean into HORIZONTAL SCALING = network effects associated with each additional unit of compute, data, or talent added Please DM us if this is you

Second, the font providing notice of the terms should stand out. Small, faint text generally won’t cut it. Hyperlinks should look obviously clickable (see if you can convince your design team to let you use blue underlines for hyperlinks).

This is an extremely well written and compelling argument for why an implicit form of risk tranching is already onchain and why a more explicit form is not needed. A must read.

DAOs 2.0 - agents can put the "A" back in DAO. But they also add teeth to "D" - b/c AI is probabilistic, not deterministic, full consensus is not required for every decision. Tokenholders can signal and the agent can compile across the collective to act.

Third, place the hyperwrap notice (“By clicking X button, you agree to the terms of service”) in close proximity to the action button where users manifest assent (the X button).

Finally, make the connection explicit: clicking the action button equals agreeing to the terms. Make the legal significance of the user’s action clear.

Can apply this same concept to make fuzzy prediction markets. “Who will be more engaging on Twitter on [date], account 1 or account 2”? Throw the inputs to the agent and let it resolve the market.

Crypto devs traditionally address this issue by using smart contracts/atomic transactions that never permit them to touch user funds I'm seeing way too many agent projects with unilateral control over user funds; projects should work with counsel to discuss relinquishing control

String it together.

We are actively looking for DAO 2.0 projects - @jessewldn compiles an amazing list of ideas here. What are we missing?

Fire design.

Thank you to @sadbryce @courtlandleer @blockaid_ @iamgingertrash @spearbit @nima_vaziri @zackbshapiro for thoughtful feedback and discussion.

12/ Hyperbolic’s founders are unmatched in expertise across Crypto & AI: - CEO @zjasper : PhD in Math (Berkeley in 2 years!), quant @ Citadel, researcher @ Ava Labs. - CTO @Yuchenj_UW : ML/distributed systems expert, PhD in CS (UW), ex-OctoAI engineer lead.

We review case law in the 9th Circuit and devise a “user control” theory to understand the CFAA. Under this theory, we conclude that if a browser extension does not control the user’s account with the platform, a CFAA claim brought against that provider is unlikely to survive

Recirculating this amazing paper from @aadams and @Mclader and others. app.uniswap.org/OnchainFX.pd… FX is an hourglass market. The opportunity for crypto to disrupt it is enormous.

If you're a founder building in stablecoins, you need to read this incredible writeup about GENIUS, STAT.

The CFAA is a federal statute originally passed to prevent hacking. Facebook, X, and LinkedIn have utilized the CFAA heavily in recent years to go after scraping services they view as a threat to their business models that rely on locking data within their platforms

(2) Cybersecurity/negligence: Control over user funds likely implicates duty of care to exert reasonable efforts to protect the funds, relevant in a negligence action

(1) Money transmission: Control over user funds is crucial to the analysis if a party is a “money transmitter” and therefore must KYC/AML (here's 40 pages I co-wrote on why: edit.financialcrimelitigator…) (here's 16 more for good measure: edit.financialcrimelitigator…)

12/ We’re deep on agents here at @variantfund so if you’re building in this space, please reach out to me and @jphackworth42. We want to hear from you.