As we keep seeing over and over again, when control of a powerful system or platform is in the hands of a few (let alone a single leader), it’s too easy to encroach on user freedoms. That’s why decentralization matters: It’s the tool that allows us to democratize systems by enabling credibly neutral, composable internet infrastructure; promoting competition and ecosystem diversity; and allowing users more choice, as well as more ownership. 

But decentralization has been hard to achieve in practice — at scale — when pitted against the efficiency and stability of centralized systems. Meanwhile, most web3 governance models have involved DAOs (decentralized autonomous organizations) that use simplified yet burdensome models for governance based on direct democracy or corporate governance — which are not designed for the sociopolitical realities of decentralized governance. However, thanks to the “living laboratory” of web3 over the past few years, more best practices for decentralization have been emerging. These include models for decentralization that can accommodate applications with richer features; and also include methods such as DAOs embracing Machiavellian principles to design more effective decentralized governance that holds leadership accountable. As such models evolve, we should soon see unprecedented levels of decentralized coordination, operational functionality, and innovations.

While it has been much-lamented, the fundamentals of user experience in crypto haven’t actually changed much since 2016. It’s still too complicated: self-custodying secret keys; connecting wallets with decentralized applications (dApps); sending signed transactions into increasingly many network endpoints; more. It’s more than we can expect users to learn in their first few minutes in a crypto app.

But now, developers are actively testing and deploying new tools that could reset frontend UX (user experience) for crypto in the year ahead. One such tool includes passkeys that simplify signing into apps and websites across a user’s devices; unlike passwords, which are more vulnerable and require manual work from users, passkeys are automatically, cryptographically generated. Other innovations include smart accounts, which make accounts themselves programmable and therefore simpler to manage; embedded wallets, which are built into an application and can therefore make onboarding frictionless; MPC (multi-party computation), which makes it easier for third parties to support signing without custodying users’ keys; advanced RPC (remote procedure call) endpoints that can recognize what users want and fill in the gaps; much more. All of these not only help web3 go more mainstream, but can make the UX better and more secure than in web2.

In the world of networks, one force invariably dominates all others: network effects. Network effects tend to be so powerful that there really are only two kinds of modularity — modularity that extends and strengthens network effects; and modularity that fragments and weakens them. In all but the rarest of cases, only the former ever makes sense, especially when it comes to open source. 

Monolithic architectures have the advantage of allowing deep integration and optimization across what would otherwise be modular boundaries, leading to greater performance… at least at first. But the biggest advantage of an open-source, modular tech stack is that it unlocks permissionless innovation; allows participants to specialize; and incentivizes more competition. We need more of that in this world.

Decentralized blockchains are a counterbalancing force to centralized AI. AI models (like in ChatGPT) can currently only be trained and operated by a handful of tech giants, since the required compute and training data are prohibitive for smaller players. But with crypto, it becomes possible to create multi-sided, global, permissionless markets where anyone can contribute — and be compensated — for contributing compute or a new dataset to the network for someone who needs it. Tapping into this long tail of resources will allow these markets to drive down the costs of AI, making it more accessible. 

But as AI revolutionizes the way we produce information — changing society, culture, politics, the economy — it also creates a world of abundant AI-generated content, including deep fakes. Crypto technology can be used here as well to open the black box; track the origin of things we see online; and much more. We also need to figure out ways to decentralize generative AI and govern it democratically, so that no one actor ends up with the power to decide for all others; web3 is the laboratory for figuring out how. Decentralized, open-source crypto networks will democratize (vs. concentrate) AI innovation, ultimately making it safer for consumers. 

In “play to earn” (P2E) games, players would often make real-world (not just virtual) money based on their time and effort playing games. This trend is related to the broader shifts that have been transforming gaming and beyond — from the rise of the creator economy to the changing relationship between people and platforms. Web3 allows us to counter the current norm where all the proceeds of playing and transacting in games go only to gaming companies. Users spend so much time in, and generate so much value for, those platforms that they deserve to be compensated, too. 

But games were not necessarily designed to be a workplace (at least, not for the majority of players). What we really need are games that are both fun to play and that also allow players to capture more of the value they produce. As such, P2E is increasingly morphing into “play and earn”, setting an important distinction between gaming and workplace. The dynamics of how resulting gaming economies are managed will continue to shift as we see P2E evolve beyond its initial growing pains. Ultimately, however, this will not be a separate trend and will just become part of games.

As someone who spends a lot of time thinking about web3 games and the future of gaming, it’s clear to me that AI agents in games must come with guarantees: that they’re based on certain models, and that those models are executed without corruption. Otherwise, games will lose integrity. 

When lore, terrain, narrative, and logic are all procedurally generated — in other words, when AI becomes the game maker — we’re going to want to know that the game maker was credibly neutral. We’re going to want to know that that world was built with guarantees. The most important thing that crypto offers is such guarantees — including the ability to understand, diagnose, and penalize when something goes wrong with AI. In this sense, “AI alignment” is really an incentive design problem, in the same way dealing with any human agent is an incentive design problem… and is what crypto is all about.

While formal methods are popular for verifying hardware systems, they are less common in software development. For most developers outside of such hard or safety-critical systems, these methods are too complex, and could add significant costs and delays. However, smart contract developers have different demands: The systems they develop handle billions of dollars; bugs would have devastating consequences, and cannot typically be hotfixed. So there’s a need for more accessible formal verification methods in software and especially smart contract development. 

This past year, we’ve seen a new wave of tools (including ours) emerge which have much better developer experience than traditional formal systems. These tools leverage the fact that smart contracts are architecturally simpler than regular software — with atomic and deterministic execution; no concurrency or exceptions; small memory footprint and little looping. The performance of these tools is also rapidly improving by leveraging recent breakthroughs in SMT solver performance (SMT solvers use complex algorithms to identify or confirm the absence of bugs in software and hardware logic). With increased adoption of formal methods-inspired tooling among developers and security specialists, we can expect the next wave of smart contract protocols to be more robust and less prone to costly hacks.

​​More and more established brands have been introducing digital assets to mainstream consumers in the form of NFTs. Starbucks, for example, has introduced a gamified loyalty program in which participants collect digital assets as they explore the company’s coffee offerings (not to mention an AR pumpkin spice maze!). Nike and Reddit, meanwhile, have developed digital collectible NFTs that they’ve explicitly marketed to a broad audience. But brands can do so much more: They can use NFTs to represent and reinforce customer identity and community affiliations; bridge physical goods and their digital representations; and even co-create new products and experiences alongside their most dedicated enthusiasts. Last year, we saw a growing trend toward inexpensive NFTs for large-scale collection as consumer goods — often managed through custodial wallets and/or “Layer 2” blockchains with correspondingly low transaction costs. Heading into 2024, many of the conditions are in place for NFTs to become ubiquitous as digital brand assets — as Steve Kaczynski and I explain in a forthcoming book — for a wide array of companies and communities.

Technologists have historically had the following strategies for verifying computational workloads: 1) re-executing the compute on a trusted machine; 2) executing compute on a machine specialized to the task, aka (TEE trusted execution environment); or 3) executing compute on credibly neutral infrastructure, like a blockchain. Each of these strategies has had limits in terms of cost or network scalability, but now, SNARKs (Succinct Non-interactive ARguments of Knowledge) are becoming more usable. SNARKs allow the computation of a “cryptographic receipt” of some compute workload by an untrusted “prover” impossible to forge: In the past computing such a receipt cost 10^9 work overhead over the original compute; recent advances are bringing this number closer to 10^6. 

SNARKs therefore become viable in situations where the initial compute provider can bear a 10^6 overhead and the clients cannot re-execute or store initial data. The use cases that result are many: Edge devices in the Internet of Things can verify upgrades. Media editing software can embed content authenticity and transformation data; while remixed memes could pay homage to initial sources. LLM inferences could include authenticity information. We could have self-verifying IRS forms, unforgeable bank audits, and many more uses that benefit consumers ahead.