⚙
Pioneer Square Labs
The Web3 ecosystem has been variously described as a collective hallucination, a massive grift, an environmental disaster, a decentralized renaissance, and the future of the Internet.
That’s a lot to live up (and down) to.
Here in the PSL Studio, our veteran engineering team (hi, nice to meet you!) has been building fun new Web3 projects. Along the way, we’ve been taking notes on what we’ve learned, what the underlying technologies do (and do not) enable, and where we see opportunities in the future.
We thought it’d be worthwhile to distill this list of observations and publish them here.
Fair warning: this is a long analysis of the current state of the ecosystem. It is a collection of opinions from a pragmatic builder's perspective. It is focused away from economics and the financial mechanisms of the chains themselves. With so much noise and churn, it’s impossible for us to predict where the cards will land. But, in this post, we promise 50% less frothy rhetoric and 100% fewer Ponzi schemes than you’ll find in a typical Web3 tweetstorm :)
🙃 First, the web is delightfully weird again (yay!)
Web3’s culture is young and vibrant. It’s reminiscent of the earliest days of the web and of many of the things that made the 90s internet fun: small communities, weird new technology, lots of blue-sky experimentation, a sense of cultural motion, the excitement of discovery, and new ways to express oneself.
Design is adventurous again. It’s technotronic; it’s blinged out; it’s brutalist; it’s Space Jam circa 2021. Money is at the center of it all, yet it’s far from templatized, corporate, or sterile. It’s a hip bar in a club that a cooler friend had to tell you about.
Web3 is home to new writing and criticism that feels like the work of whip-smart semiotics majors. Like the web’s earliest days, there’s plenty of lofty hyperbole about crypto’s utopian potential. New social clubs seem as enthusiastic about throwing underground raves in Paris as they are with the market value of their tokens.
Sites are admirably befuddling again. Nobody quite understands what they’re building, let alone how to describe it to you. But if you click enough (and maybe connect with MetaMask), you’re equally likely to end up with confusion as you are with an amusing or revelatory new experience.
Ignoring all the technical details and religious wars about Web3, it’s just plain exciting to watch as the web’s new generation emerges. But… there are some technical details to discuss. All of this new energy is built upon bona fide new technologies. Let’s talk about those first.
⛓ Web3 is inseparable from blockchains and cryptocurrencies
“Web3” is the name given to a suite of peer-to-peer technologies — particularly blockchains and distributed filesystems — that are used to build modern “decentralized apps”, or dApps.
Blockchains are databases built from three parts:
It’s expensive to participate; incentives are necessarily financial. At the same time, blockchains are an ideal data structure for managing trusted ledgers. Thus, cryptocurrencies and blockchains are inseparable; they’re flip sides of the same coin. (Sorry.)
🌎 Blockchains flip the script on Internet protocol development
The history of the Internet is, in part, the history of the birth, adoption, and stewardship of distributed protocols by the broader community. Blockchains follow in this tradition, but also break radically from it: they are the first protocols to arrive with an asset class attached.
Protocols like SMTP (1981; email), TCP (1983; reliable packet transmission), HTTP (1991; web), and XMPP (1999; chat) all created immense value while capturing little for their inventors. Blockchains upend this, allowing inventors to capture considerable value for themselves. (See also: “Fat Protocols” by Joel Monegro.)
Historically successful Internet protocols typically see slow early adoption, gather steam over a decade or more, and demonstrate increasing value as they settle into stable long-term use. Blockchains — and also the services built on top of them — break this mold by nearly requiring up-front investment (or wild speculation!) to achieve escape velocity. This leads to a problematic dynamic: legitimate projects share many structural characteristics of Ponzi schemes or multi-level marketing. These schemes could well ultimately collapse; many early-stage blockchain projects probably will. But it’s not a certainty: if a protocol’s underlying economics are sound and there’s demand for the product, what might look like a Ponzi scheme at first could be the “go-to-market strategy” for a network that creates durable long-term value.
It remains to be seen how the massive amount of capital behind blockchain-based projects impacts their longevity and growth. It also remains to be seen whether they can or will create value commensurate with their considerable market caps ($2T+ collectively).
💫 Blockchains are not equal; programmable blockchains are rich and strange
Bitcoin (launched 2009; $1T market cap) introduced the world to modern distributed consensus and laid the foundation for all that has followed, but in critical respects Bitcoin itself is quite limited: to a first approximation, it’s simply a ledger that tracks bitcoin ownership. There’s a pleasing circular simplicity to the system.
Ethereum (2015; $400B) is a different beast entirely: rather than maintaining a simple ledger, Ethereum’s blockchain is used to maintain the state of a virtual machine — a single fully programmable computer shared by everyone on the Internet. By writing and deploying “smart contracts” — code intended to run on the Ethereum Virtual Machine (EVM) — developers can guide the future evolution of the system. This has given rise to new digital constructs like NFTs and DAOs.
Solana (2018; $50B) is one of several programmable chains that followed in Ethereum’s wake. Solana’s focus is on enabling high-throughput low-cost transactions, two crippling weaknesses with the current Ethereum network. Several innovations in distributed consensus design make Solana’s performance possible. A16z led a $314M investment in a recent Solana Labs token sale.
🤨 Very few categories of data belong on-chain
Is the blockchain right for your data? Probably not. There are many considerations, including:
A good rule of thumb: unless (1) you need to shout to the world that action A took place at time T, and (2) the world cares enough to pay attention, you don’t need a blockchain*. Updates to financial ledgers are a perfect example; despite a decade of optimism from enthusiasts, few others have been found that have achieved broad adoption.
* Just because you don’t need a blockchain, it doesn’t mean there won’t be enormously successful applications built for that use case on the blockchain. See “You don’t need a blockchain to do X” below.
🪙 Ledgers, coins, and tokens are the abstractions upon which all crypto is built
As described above, blockchains are great for maintaining ledgers: simple lists of who owns what.
The “who” is an account. In the case of programmable blockchains, that’s either a person holding a private key, or it’s a smart contract.
The “what” is either a coin or a token. The distinction has somewhat fuzzy boundaries but, roughly speaking, a coin is a blockchain’s intrinsic currency, like Ether, Bitcoin, Sol, or Dogecoin. A token is an asset defined on top of a programmable blockchain, like Shiba Inus or CryptoPunk #1772. (A naming exception is made for “stablecoins,” which are closer to tokens in implementation.)
Everything else in the crypto ecosystem layers on top of these central abstractions.
🐵 Blockchains allow anyone to assert the existence of new digital assets
Programmable blockchains make it so that anyone can wave a “magic wand” and create new digital assets.
Tokens are abstract assets, and they come in two flavors. Fungible tokens are mediums of exchange, whose supply is controlled by code. Non-fungible tokens are unique assets, potentially drawn from a class of related assets. By construction, blockchain users must honor these creations.
Of course, simply waving crypto’s “magic wand” doesn’t create value. It’s everything that’s wrapped around the tokens that matters. Early on, the community experimented by attaching pictures of punks and apes to its tokens. Later, it tried stapling complex structured metadata to them, for example for NFT game characters. Today, we’re starting to see sophisticated code wrapped around tokens: code that grants permissions and access rights, code that defines and enforces behavior, and code that describes how tokens interact with crypto’s new payment rails. This is a vast design space to explore. If the crypto economy has productive ends, they must ultimately rest here.
It’s worth commenting on the reach of crypto’s “magic wand”. Tokens exist on blockchains; it’s only if other software interoperates with those chains that the tokens matter. The entire crypto ecosystem is a collective act of agreement to honor the abstractions in these weird databases. There’s no fundamental reason the whole world should agree to this. There’s also no limit on the number of potential chains on which people might wish to declare the existence of new digital assets. Where this leads is anyone’s guess. It’s unlikely to be the universal fully interoperable ecosystem of digital assets that some imagine. Anyone who is playing around in DeFi in both the Solana and Ethereum ecosystems today can attest to that!
😳 Today’s smart contract programming models are deeply flawed
Modern programmable blockchains like Ethereum and Solana are impressive technical achievements. At the same time, smart contracts have some deeply problematic constraints:
These are serious issues that will need to be addressed. They are potentially interesting opportunities for dev and infra tooling. We expect to see substantial innovation going forward as a result of these nontrivial problems. (We have some guesses about how each of these might shake out over time; drop us a line if you’re thinking about solutions in this space — crypto@psl.com.)
🧮 Distributed consensus technology could change radically in the next decade
As described above, today’s blockchains allow users to “shout things to the world”. To achieve this, the entire history of the chain must be replicated by each full participating node in the network. This is expensive and slow even today for six-year-old Ethereum. Worse: it doesn’t match the real-world. Even when people need to “shout” things to one another, they often don’t need to shout them to the world. In theory, a private party transaction should just involve the two parties; today on the blockchain, it needs to involve the entire world.
New classes of zero knowledge proof systems under study at the Stanford Center for Blockchain Research (amongst several other academic institutions) may make it possible for private parties to share just the necessary bits of economic history and fact with one another, without the need to consult a central blockchain ledger. If practical, these new approaches could radically alter crypto’s technology landscape and fundamentally upend current valuations. While these new approaches are promising, at the moment they’re firmly in the realm of theory, not practice.
🌌 We should expect to live in a multiple-blockchain world, with all its complexities
Quite a few blockchains have $10B+ market caps today (over 20!); it is hard to imagine that most of them will go away tomorrow. Each blockchain defines its own unique ledgers, and therefore its own unique digital assets.
Like anything distributed on the Internet, there is likely to be some amount of federation between blockchains and some amount of true disconnect between them. The decision to connect two blockchains, or to keep them disconnected, has strategic, economic, and technical components.
There are several instances of connection today. Take for example the Solana-to-Ethereum wormhole, which allows tokens to be created on the Ethereum blockchain that represent assets on the Solana chain, and vice versa. It is a technically sophisticated construction: the smart contract deployed at one side of the wormhole needs to “own” the assets transported to the other side; the smart contract on the other needs to mint new “wrapped” tokens that act like remote proxies for the real things. It’s also a rickety construction: some code that sits outside of the wormhole contracts needs to transit data between both ends since, of course, smart contracts can’t reference smart contracts on different blockchains!
😬 Blockchains are regularly credited with powers they do not, in fact, have
We’d like to take this moment to politely disagree with several common bits of rhetoric we feel we regularly see about blockchains:
🏘 One way to think of tokens is as pure abstractions of potential rights
Chris Dixon claimed in a recent tweetstorm that “tokens give users property rights: the ability to own a piece of the internet”.
While this sounds like it might be true… today, they absolutely do not. But, squinting a bit, there’s no reason they couldn’t tomorrow: a reasonable future framing of (non-fungible) tokens might be that they are pure digital abstractions to which a bundle of property rights can be attached.
Of course, to arrive at this very theoretical future, we’d need public/private partnerships to bridge the world of crypto with the normal legal and regulatory world. This road to this potential future is long and winding. Whether it is at all desirable to follow this road is left as an exercise for the reader.
🔮 “You don’t need a blockchain to do X,” even when true, might miss the bigger picture
A common retort amongst Web3 skeptics (and by ourselves, earlier in this piece!) is that, for a given use case, blockchains are “databases but worse”. And… the skeptics are often right!
However: just because you don’t need a blockchain to do something doesn’t mean the industry won’t settle on using blockchains to do it anyway. The technology industry is immensely path dependent. Particularly when buckets of money appear, feedback loops can form whose outcomes seem all but inevitable. Speculators and venture capitalists alike have inundated crypto with cash. There’s a certain reflexivity to it: when cash pours into anything, the intrinsic value of the thing is at least the value of the cash… and potentially much more, if that cash is invested productively! Blockchains may become the future — including for use cases where they are not natural fits — only because their story was told, speculated on, invested in, and told some more.
📈 But crypto’s current prices are… big numbers!
Plenty of ink has been spilled on crypto’s outsized valuations.
There is no doubt that we’re living in a speculative bubble of immense scale. If you invested $1,000 in Shiba Inu tokens on March 1, 2021, you’d have had $2.815M in early October; the pups would have barely grown! This same story repeats itself throughout the ecosystem: that 300% APY you get from staking SUSHI tokens in a yield farm contract is a mirage that can only last so long.
One observation that we’ll offer on crypto’s gigantic market caps: while valuations of traditional corporations (especially technology companies) have skyrocketed in recent years, they are still ultimately grounded by well-understood, time-tested business fundamentals. Even if a business is worth twice the revenue multiple you would have assigned it 3 years ago, there is still some way to tie it to a well-understood metric. In crypto, the systems work differently. Value accrues to different parts of the ecosystem, like the users or the community. Often there isn’t even a company at all! Adoption can happen so quickly (especially since users are incentivized to adopt) that it further encourages speculation. The lack of standardization on how to derive valuation, coupled with the Wild West of new mechanics, creates the perfect conditions for crypto’s crazy high valuations to seemingly emerge overnight.
☠️ “But crypto is used for illicit stuff!”
There are even more problematic use cases for crypto than rampant speculation. In many ways, the world of crypto contains aspects of an unregulated casino. Big-time punters, grifters, money launderers, and ransomware bandits are no doubt hard at work inside this casino. This may change over time, but not without considerable regulatory oversight.
Nobody who spends time or builds in this ecosystem should ignore this raw truth; those who build best will constantly consider it.
🚜 Crypto is a (well-funded!) random walk towards discovering productive economic ends
Ultimately, a healthy economy needs productive ends. It remains to be seen whether crypto, with all its capabilities and limitations, has productive ends. Luckily, the ecosystem is awash in capital and has a culture of rapid experimentation; it may well uncover a pot of gold.
If anything, crypto’s productive ends will have to be found in the tokens it creates: it’s the digital assets themselves that must be the underlying source of value.
As mentioned above, the design space for tokens is vast and largely unexplored as of today. The space covers how tokens intersect with access, ownership, and intellectual property rights. It includes how tokens align incentives between disparate parties. It spans the decentralized reinvention of, seemingly, all of modern finance.
🚷 Onboarding to crypto is easy; going deep is walking on fire
The typical user’s first foray into the crypto ecosystem is through fiat on-ramps like Coinbase. Coinbase maintains a wallet* on behalf of its users. The experience for users is much like the typical web: log in with a username and password, get access to your investments.
To wade deeper into the web3 world, users need to install noncustodial wallets like MetaMask: software tools that are functionally “1Password but for your crypto private keys”. Unlike 1Password, however, MetaMask is a user experience disaster from its introduction (the website does an impressively poor job of describing what MetaMask is, or why you might need it) to its daily use (strange layout of common controls, lots of weird technical stuff like hex addresses very much front-and-center, no understanding or presentation of NFT ownership, etc). If 1Password’s users make a mistake, they might reset their passwords via email. If MetaMask’s users make a mistake, they might lose their investments forever. And MetaMask isn’t all: sophisticated users need multiple crypto wallets (Phantom, etc.) since they work on multiple blockchains.
Plenty of opportunity exists to make better wallets. It is a remarkable testament to the demand for noncustodial wallets and the use cases they unlock that, despite all this, MetaMask has over 10 million monthly active users!
Companies like Brave have taken a different approach to integration with crypto, making it a top-level feature of their web browser. While still a small user base by browser standards (still under 1% of global market share), Brave has demonstrated impressive traction, with over 40 million monthly active users.
Beyond wallets and browsers: the typical user-facing Web3 service is built for those “in the know”. Visit OpenSea and you won’t see much by way of explanation about what’s going on; visit Mirror.xyz and you’ll see even less. And yet these are interesting, deep services for those who know what they do and how to use them.
* (Coinbase’s wallet is really several wallets, since Coinbase interacts with multiple blockchains. This is entirely hidden from Coinbase’s users — a good thing for gaining the most mainstream usage, but also a true stumbling block for going deeper.)
💰 Coinbase and Robinhood could quickly centralize that which isn’t
As of October 2021, Robinhood appears to have ~10M active crypto users; Coinbase appears to have ~50M users. MetaMask claims to have just surpassed 10M MAUs, so it’s about 1/6th the size of the crypto trading ecosystem. All of these services are growing rapidly, but our read of the (admittedly sketchy) available data points is that the number of crypto users with custodial (cloud-based and centralized) wallets is growing meaningfully faster than that of noncustodial wallets; growth in the former isn’t necessarily translating to conversion to the latter.
Access to NFT purchases and more obscure DeFi products seems to be the primary driver behind MetaMask’s growth. With the introduction of Coinbase NFT, it’s clear that Coinbase wants to own this growth directly. Perhaps it won’t be long before Coinbase builds direct interfaces to the more popular DeFi projects.
Further: Coinbase already offers its own custom API that developers can use in addition to — or instead of — interfacing with noncustodial wallets. It hasn’t seen immense adoption, but there’s plenty of opportunity for future growth. For example, Coinbase could offer the ability for developers to build subscription billing services in a way that’s simply not possible with the blockchain itself (because every transaction must be separately signed).
The promise of decentralization on the internet was empty the first time around. Somehow, what started as a decentralized network of servers and open protocols became a concentrated powerful set of a few trillion-dollar companies. Web3 may be different, but the question remains open. Stepping back: economic power always concentrates; power laws are the norm. Web3 may mint new winners, but the distribution of returns may well stay the same.
🗺 DAOs explode of the idea of “company” into infinite forms
There are two roughly equivalent ways to describe Decentralized Autonomous Organizations. First, they’re smart contracts that interact with and coordinate people at some level. Second, they’re groups of people in crypto-land working towards a common goal, whose actions are partially (or fully) mediated by code.
Nearly all DAOs share two things in common: (1) a governance token, which grants voting rights to its holders, and (2) a system for proposing actions and voting them up or down.
Beyond this, however, the “goals of groups of people” covers an impressively wide range. Some DAOs have multiple token classes; some only one. Some DAOs tie their votes directly to action via smart contract code; most don’t. There’s a huge taxonomy of DAOs emerging in the wild, all different:
With so many DAOs in the world, we’ve also seen the genesis of a modular ecosystem of services designed to help people build and manage yet more DAOs. It’s still very early; aside from tools for managing voting (like snapshot.org) and an increasing collection of common smart contract code (see OpenZeppelin’s Governance contracts), there simply isn’t a clear locus of activity here yet. The ambiguous legal and tax status of DAOs probably further hinders adoption.
💸 Decentralized Finance (DeFi) is an explosive corner of the crypto world
In a telegram thread in August 2018, “DeFi” was coined. It quickly became the umbrella term for a rapidly-emerging set of peer-to-peer financial services on public blockchains. DeFi enthusiasts will take anything that a centralized bank can do — lend, borrow, trade assets, trade derivatives, etc. — and enable the same sort of interactions in a peer-to-peer system with no single point of failure.
Ethereum, with its fully programmable underpinnings, proved to be the blockchain in the perfect place at the perfect time for early DeFi projects to flourish. This led to a flywheel effect: Ethereum enabled the proliferation of early DeFi services, and DeFi massively increased the usage, market cap, and (unfortunately!) gas fees of the Ethereum network.
Over time, Ethereum has proven to be sub-optimal for the incredible demand DeFi services have seen. The network has diseconomies of scale and, given its low throughput capability, gas fees (the cost of executing a transaction on the Ethereum network) have recently been over $150. This makes executing simple low-value transactions, such as swaps on Uniswap, prohibitively expensive! Since the emergence of DeFi, there are several other L1 chains (like Solana and Avalanche) that have become a popular place for DeFi developers to build with transaction fees. Ethereum itself is also in the midst of the infamous “ETH2” transition, which aims to solve this problem on the Ethereum network itself.
🧠 DeFi services marry economic insights with clever code
Even if today they’re only good for speculation (and boy are they ever used for speculation today… often with an incredible amount of leverage), there’s quite a lot to learn from the design and implementation of decentralized financial services like MakerDAO, Uniswap, and Compound Finance.
MakerDAO is a DeFi building block that attempts to create a “stablecoin” — a token (DAI) pegged 1:1 to the US Dollar — through collateralization. The basic approach is straightforward and starts with the realization that, while the prices of Ether and of popular tokens in the Ethereum network are quite volatile, there’s a likely limit to that volatility. ETH is worth ~$4,000 USD today; it might be worth $2,000 tomorrow but it’s probably never going to go so low as (say) $1,000 in the future. By putting up 4 ETH, MakerDAO can manufacture 1 DAI and have reasonable confidence that it can maintain par value. The DAO itself, through membership with the MKR token, votes and regularly updates the underlying economic parameters: the set of tokens accepted as collateral, the required ratios, and payouts when underlying asset prices rise.
Uniswap is an automated market maker (AMM) that implements a decentralized exchange (DEX). Users can swap one type of token for another at some economically sensible ratio. The fundamental job of a traditional exchange is to maintain an order book that matches buyers to sellers. In the decentralized world, there’s no third party to maintain that book. Instead, Uniswap incentivises participants to contribute to “liquidity pools” — pairs of tokens (like, for example, SHIB and SUSHI) that other users may want to exchange. Liquidity providers receive payments for the tokens they’ve staked (via “LP tokens”); users can get the specific tokens they desire by exchanging the tokens they have. Each liquidity pool has an associated exchange curve (the “constant function” curve); the curve is parameterized based in part on the ratio of tokens on either side of the pool. Uniswap v3 substantially increases the flexibility of the service, allowing liquidity providers to stake at “concentrated” positions along the exchange curve. Tie all this in with a protocol DAO (via the UNI token) that manages economic parameters, and you have a truly innovative bit of code and finance.
Compound Finance bills itself as a decentralized “interest rate market”; it was the first such in the crypto world. The economic realization is that there’s an opportunity to fill the gap between people with surplus assets they can’t use, and people who could use those assets today; Compound effectively allows trades on the time value of those assets. Somewhat like a traditional bank, lenders contribute tokens to the system, and borrowers can decide whether to borrow at current rates. The underlying economic parameters of the system are managed by the holders of the COMP governance token.
💵 There’s infinitely more to say about the financial side of things (that we won’t delve into here)
If you’re an ambitious quant hanging out at a high frequency trading firm, crypto has got to be nearly irresistible. It’s an alternate universe with immense cash inflows (let’s use the amount of stablecoin issued today as an absolute lower bound, so maybe $150B… or we could use Sam Bankman-Fried’s recent estimate of $400-500B) and demand, coupled with a weird and very young new settlement layer. Clearly there’s gold in these hills — perhaps it’s a gold rush that will peter out, or perhaps it’s a once-in-a-lifetime opportunity to rebuild the financial system from the ground up.
On this point, Sam Bankman-Fried has a pragmatic and clear-eyed take on how the financial world wants to build in crypto-land. He’s pretty clear that smart traders who understand market structure and dynamics are quite excited to sit on the other side of retail activity in the space — and he’s built FTX with that in mind. (See, for instance, this episode of the “Invest Like the Best” Podcast.)
As an aside, it’s worth noting that SBF built Serum on top of Solana. He did it for the simple reason that transaction throughput, latency, and cost really matter to his customers — they come from the world of traditional finance, where systems operate much faster.
👩🏻💻 Blockchains are the kinds of databases that frustrate developers
Blockchains are databases, yes, but they’re not databases designed for traditional online analytics processing: they’re not designed to be easily queryable.
An entire cottage industry has emerged around building blockchain ETL systems: specialized code extracts relevant information from the target blockchain, and then loads it into either a SQL database, or into a data warehouse that offers some kind of rapid query functionality. A number of open source tools exist, including the young but still-quite-useful ethereumetl project.
Extracting usable data from blockchains is hard. It’s hard because there’s a lot of data; extracting the data rapidly, at-scale, and at-cost is by itself a nontrivial challenge. It’s hard because there’s always new data: a great blockchain data system will need to both batch and stream data. But more than that: it’s hard because the data on programmable blockchains, in particular, requires interpretation based on the actual smart contracts deployed, and on the ecosystem surrounding those smart contracts.
A few early businesses have emerged around working with blockchain data. For us, the two most notable are dune.xyz and nansen.ai. They’re notable because they are (in our opinion) the two most technically advanced projects, and they have found very different go-to-market routes.
Broadly: there are many more developer tools that are worth mentioning here. The developer tools space is very nascent. The industry is ripe with opportunities to improve the developer experience, including but by no means limited to blockchain data.
💘⚰️💘 Some parting shots
If you have any thoughts on this piece, or are building in this area and would like to talk with us, drop us a line! crypto@psl.com
You can also discuss this piece on Twitter.
Archera raises a $7 million Series A Round
Our Investment in Starfish Space
Fortive Acquires TeamSense - A Major Success in Corporate Innovation
Our Four Questions to Validate an AI/ML Startup Idea
Launching PSL Ventures II, a $100 Million Seed Fund for the Pacific Northwest
Introducing: Tech Jobs Demystified
Guest Post: Diversity from Day One at TeamSense
What's The Big Deal About GPT-3 (And What Does It Mean For Startups)?
Our Investment in Copper Banking
The Tools We Built to De-Risk Startups Faster
What we’ve learned about validating startup ideas during a crisis
The Software Stacks Our Early-Stage Companies Are Using in 2020
State of the State of the Art in ML and AI — Q4 2019
Our Investment in Sentinel Healthcare
Our Investment in Super.AI (formerly Canotic)