Are Protocols Defensible in the Age of AI?
A must-answer prompt for protocol developers.
The more I read and think about AI, I'm able to convince myself that I'm gaining some clarity on where enterprise value will accrue in existing markets.
But I find the same question very difficult to answer for crypto.
On one hand, crypto by definition is perfectly suited to subsume the set of technologies that require no interaction in “meatspace”.
On the other, today’s successful crypto applications are de-correlated from the industries that seem to be transforming most quickly: content production, services, research, B2B SaaS, etc.
I'm very confident that new protocols will be able to attract profits going forward (we’re building one), but it’s not obvious which ones will be able to retain it.
The question is simple: how do language models impact protocol defensibility in crypto?
TRADITIONAL DEFENSIBILITY
I still prefer the 7 Powers lens as a way of looking at defensibility, but some of its factors are shifting in relevance.
Counter-Positioning ++. Doing things that bigger competitors won't do to avoid self-cannibalization is still an important strategy to compete with entrenched brands. May become less necessary to compete with more value based products as innovation speeds up. I'm not sure.
Cornered Resources +. Cornered resources are still important but their scope is becoming more narrow. Some resources won't be valuable anymore, others become cheaper and cheaper to reacquire (e.g., research reports). I do think on balance the “data is a moat” argument may have a little more legs than it did before although I'm still not its biggest fan.
Scale Economics +++. Seems to matter more now? Commoditized intelligence makes compute more valuable. This is what Shravan Narayen had to say about it:
Switching Costs +. Switching costs will likely see the biggest disruption. They are not going away but it’s important to examine who is switching when the way we interact with tools will be completely different. There may be CRM tools that you use without even logging in because you are accessing them from your agent.
Network Economics ++. Likely still an important factor, however, it does depend on how users extract the value from their network. Networks that reflect friend or para-social relationships are going to become stronger than interest graphs as the way we curate information changes.
Process Power +++. This is the one to bet on. Unlike every other technology, I find it difficult to imagine that the development of agents works on an S-curve. I'm just not convinced that the returns to higher intelligence, information-gathering velocity, ability to self-replicate are sub-linear. So if the returns on intelligence, learning, replication are super-linear and can help acquire any of the other powers, process-power or the ability to continue uncovering new unmet customer needs and developing targeted AI workflows to service those needs may become the key differentiator.
Brand +++. All important. Strong brands will see the least immediate disruption from AI. You don't really forget a brand you already trust just because another product has come along although brands do risk becoming outdated when their products do. AI may accelerate the creation of new brands, however.
WHY PROTOCOLS WERE DEFENSIBLE
It’s hard to look at protocols directly in isolation from the front-ends and companies that support them. The framing I prefer is looking at the stack surrounding the protocol as part of the moat.
For protocols only developed by one team it’s more helpful to look at the moats of the whole product suite in conjunction, but lets consider the more general case here.
One of the primary moats that protocols have is that they were developed by a team (cornered resource?) that is incentivized to see the protocol succeed.
The Uniswap protocol benefits from the Uniswap Labs front-end that routes LP activity.
The success of Synthetix is at the hands of Infinex.
And AAVE is stronger because of Avara.
The other common moat of protocols is liquidity network effects.
Commonly applicable to study AMMs, more liquidity lends to more issuance and trading, which in turns lends to more liquidity.
The problem with liquidity moats is while they do reinforce advantages when the protocol is “fit for purpose”, they don’t in itself incentivize protocol upgrades.
Zora didn't become a Hyperstructure because it had to keep evolving. It’s now exploring fungible issuance.
Even some of the most successful products like Uniswap v3 and AAVE have completed or are planning to complete holistic upgrades.
Uniswap v3’s flywheel will likely allow liquidity to persist for a while but at some point it will be completely flipped by Uniswap v4 which enjoys the same compounding benefits but is a strictly superior superset and also has a hook developer flywheel.
So it's questionable whether the full network effect is truly tied to the “protocol layer” as opposed to at a community level.
In fact, referring to the image of this post, protocol moats seem to exist at an equilibrium between growing the power and quality of a protocol’s stakeholders (e.g., token holders, delegates) who create integrations and positive marketing and growing the network effects of the direct protocol.
However, there is an important exception to the limitations around protocol flexibility: tokens.
Well-adopted assets like stETH have much stronger staying power in crypto than permissionless protocols because they universally integrate with most apps and other protocols that work with ERC20 making their network effects far more entrenched.
(Protocols always have room for asset issuance as Figue from Paladin has pointed out in regards to protocol stablecoins.)
In some ways protocols these days spin up several “adjacencies” to recapture value or surrounding network effects:
Tokens (recapture monetary premium and liquidity network effects)
Front-ends (recapture solver fees)
L2s (recapture priority fees and developers)
While the competition in L2s and front-ends is ruthless, monetary premium for protocol-native tokens (governance or not) has persisted well.
EVOLVING ROLE OF PROTOCOLS
It’s undeniable that while protocols exist as pieces of code, their designs are very tied to how humans operate outside of the blockchain.
First, protocols have become much more solver friendly over time, supporting account abstraction, gasless transactions,
Permit2, solver-based settlement, etc.Second, protocol operations rely on off-chain intelligence more than ever with the rise of curators.
Third, new-generation protocols are careful to delegate too much active development to governance trying to find ways to be more efficient by spinning up Foundations or SubDAOs.
Protocols don't just exist “in the matrix”, they are a reflection of the real world.
And if that real world is rapidly impacted by AI development, protocols may have to adapt too.
We can consider a number of lenses for what crypto provides to the world to see how protocol defensibility will need to evolve.
Lens 1: Protocols replace financial rails.
The Paradigm TradFi report suggests that we've already built a better platform for global finance where broader adoption is blocked largely by regulation.
This regulation-centric view of adoption suggests that blockchains are well positioned to have staying power as long as current lobbying efforts and momentum persist.
Lens 2: Humans control AI & blockchains become a substrate
Networks of agents will need a platform to coordinate and exchange value. Blockchains work for this, providing purely cryptographic assets, support for arbitrary smart contracts to govern agreements between agents, etc.
In an optimistic world where agents get better but are still controlled by humans, humans operators could continue to use existing blockchains and protocols they know, understand and trust to allow agents to interoperate.
This is not just something that crypto-anarchists understand, from Ben Thopmson:
Lens 3: Agents control their tech stack
It’s also possible that agents could coordinate to build better-purpose tech stacks.
You don't need anything but code and node operators to “spin up” a blockchain. In this world, current blockchains are just an initial blueprint for what an optimized agent medium could look like.
In this version of the world both existing blockchains and protocols would see significant disruption.
This shift would also accelerate protocol design iteration making other moats much less pronounced.
BUILDING FOR THE FUTURE
While there is a lot of uncertainty, there are two clear takeaways for protocol developers.
Respect the changes happening off chain.
To have staying power, protocols have to become adaptive to how quickly the world is evolving outside crypto.
Two examples:
Uniswap v4 benefits from better AMM design patterns as they can be used to build better and better hooks. It will likely remain the fastest path to order flow to any vibe coder using Cursor or Windsurf to build a new AMM.
Aera benefits from better off chain strategies as they can be used to build more sophisticated guardians. It is entirely compatible with a world where the best guardians are AI agents too. This is true for other curated protocols as well.
Issue assets
Assets may be the best “store of power” that protocols have available. It allows protocols to leverage whatever advantages they have today into a preservable network effect centered around a token.
Protocols with existing advantages should seek to issue and develop assets not just to add additional revenue but make their community more asset-centric as opposed to protocol-centric.
It may have been easy to conclude that we are entering a post-defensibility market but that is simply not true.
Static moats are becoming weaker and choosing and betting on the correct futures requires careful consideration.
A must read for those in the industry!