In the first part of this essay, I explored what specifically a DAO is, and introduced two ideas: that they automate middlemen in banking and administration. But I also spotted a problem, which is that trying to build an organisation to take advantage of both of these innovations can cause difficulty. That’s because automating the banking aspect lends itself to expert partnerships, whilst automating the latter lends itself to expansive community, which actively undermine each other.
Trying to do both at the same time causes inevitable tensions. The next two pieces in the series are about mitigating those tensions, and this month, I’m focusing on how DAOs built towards expert partnerships with centralised holders can navigate the tensions that arise as they attempt to embrace their community.
Such centralised holders may have 20% of the votes, but they have far more important things to worry about than spending such a large share of their time on the DAO. We can see these DAOs appear in the top-right corner of last month’s diagram, where the top-left DAOs are built around expansive communities, and the bottom-right DAOs are built around expert partnership.
To be clear, the top-right corner is not a ‘wrong’ corner to be in. DAOs in the two goldilocks zones will have fewer ongoing governance tensions, but face limitations in the activities they can undertake effectively. By contrast, it can be a sensible decision to form a DAO in the other quadrants too, where your organisation can have a larger scope.
One common group of organisations in this top-right zone are consumer-facing services, like Uniswap, Compound, and Lido, providing financial services that could scale to millions. These organisations benefit from the technocratic decisions of a core centralised leadership, but when at this stage, they can also benefit by decentralising some authority to users.
Where does this leave the original holders? Venture capital funds are an extreme example: when Andreessen Horowitz owns tens of millions of dollars-worth of Uniswap tokens, they have enormous influence in the DAO, but this stake means little to them compared to their full asset portfolio. It’s no surprise that the venture capital fund has delegated their tokens in Uniswap and other DAOs (dYdX & Fei) to university research groups. Active governance is not worth their time. (Meanwhile, a16z benefit by getting access to the world’s up-and-coming web3 talent).
It’s not just companies. Kenneth Ng and Jesse Walden, who have both worked with Uniswap for years, stand out as major holders who are not as involved in Uniswap as their influence suggests they should be:
Kenneth used to run the Ethereum Foundation, where he subsequently helped Uniswap receive a grant. His tokens are delegated to Boris Stanic, a governance contributor who leads a Uniswap grantmaking sub-committee and makes votes for Kenneth on his behalf.
Jesse’s tokens aren’t delegated, but he’s participated in far fewer votes than everyone else (whichever way you cut it). He’s also just raised $450m to invest in web3 via his venture fund, so I think he’s too busy for Uniswap governance.
Looking more broadly secures Uniswap’s status as a top-right quadrant DAO. Just 60% of their tokens are reserved for the 1.5m people who use the product, with the rest distributed to team members and venture funds. Just nine wallets currently control a majority of the votes.
This necessarily implies tension because the community half comprises users who became owners, and half owners whose small stake far outweighs any gain they see from being users. High-influence holders, low-influence users.
And the thing with owners is that they are constantly weighing up the choice between growing the organisation, which benefits users, or extracting from users, which doesn’t.
For Uniswap, this issue came to the fore last year, when one of the groups to which a16z had delegated votes, Harvard’s Law Society (HLS), pushed through a vote to donate $20m to itself to fund education around web3 finance. The vote, which eventually finished 43m to 13m, was only approved thanks to 35.5m of votes from this core a16z-Ng-Walden bloc: 10m from HLS itself, 7.5m from two other university groups voting with a16z tokens (UCLA & Columbia), and 18m votes from Kenneth and Jesse. The vote would have lost 7.5m to 13m without them.
With concentrated holders calling the shots, this naturally raised the ire of the nominally decentralised community. But what frustrated the community most was the limited accountability that this implied. HLS had promised to sell the tokens over a multi-year period, so as not to depress the price for everyone else. Yet, they went ahead and did just that within weeks, selling $10.2m in one go.
There are genuine reasons to do this, like volatility in crypto markets, but the point is the unilateral way in which it was decided: conversations between high-influence holders that left low-influence users confused and bitter. Other Internet, a research group, argues that this division manifests much more broadly, across political, social, and economic areas:
Political: the tension governance institutions face to be accessible to all whilst remaining efficient
Social: the tension between an autonomous or directed community
Economic: treating users as customers or stakeholders (the extract or grow decision)
Elsewhere, they observe that this ultimately stems from the concentration of governance votes. This blurs the lines on where the authority of the founding team ends and where the community’s authority begins. I agree: this is a problem inherent for organisations which try to build mass community ownership into high influence expertise-driven partnership.
This problem can be mitigated, however, by governance institutions which demarcate those lines of authority. That said, to design those institutions appropriately, we first need the right understanding of the problem at a deeper level.
Specifically, we need to ask ‘in what way should we demarcate the lines of authority to get the best out of the DAO'?’
And some may notice that there is an even broader claim implied within this one: that it is even possible to successfully design these institutions before creating the DAO.
In a recent blog post, Vitalik Buterin accepts this broader claim and attempts to apply the notion of concave and convex decisions to the lines of authority question. These are uncommon words for simple ideas. Concave decisions are when the best decision is a compromise between two options; convex decisions are when the compromise would be worse than either option.
I don’t think this makes sense. At first glance, it might be natural to believe that communities should have control over concave decisions (that a community vote will naturally lead to some middle ground), whilst expert partners should have control over decisions where there is a need to pick between options that don’t mix well.
Yet this conclusion relies on quite a few assumptions. Specifically, it assumes that options A and B lie where they are. If option B was in the middle, at where the curves asymptote, then some experts would be better than compromise, and some would be worse.
But more broadly, it assumes that we have some knowledge about what these curves look like before we have a vote. This defeats the point: if we knew what the curve looked like, we would know the best approach anyway. But we do not, and we cannot guarantee our ability to identify whether a specific decision is concave or convex.
Vitalik notes this, but suggests that we can still come out ahead by putting certain types of decisions to concave or convex decision processes. But it is just as hard to know whether identifiable types of decisions are concave or convex, making it difficult to build a governance structure around this idea. Decisions that we may give to communities might be better when decided by technocrats, and vice-versa, whilst these communities still may make a good decision, but only because the options with which they were presented happened to fall at certain points on the curve in ways that we misunderstood. Doing governance this way is like the blind leading the blind.
If there is an overarching rule for where community governance should be involved, I do not think the concave/convex model is it. I think the best that we can do is the distinction between strategic and operational decisions. These two different types are not based, as above, on whether compromise helps or hinders. Rather, this model is based on what it takes to get the best answer.
In this way, it is similar to two overarching modes of innovation: incremental or radical. Incremental innovation occurs when innovators with deep expertise continue to add incremental improvements over the long term. These are our operational decisions, which requires deep understanding of how a system works before you can successfully execute a plan to achieve your goals.
On the other hand, radical innovation occurs in new industries, creating new approaches where connecting lots of dots, across breadth rather than depth, is best. This is what communities offer, the ability to synthesise a large spectrum of views to inform strategy, vision, and market decisions. In other words, community members are typically less able (and less keen!) to make specific decisions about how grant money and other resources should be allocated, but are well-placed to establish guidelines and principles to govern those allocation decisions and elect accountable representatives to implement them.
Decentralisation trade-offs are not new, and we can learn from those who have made them before us. Johnson & Johnson, now a conglomerate of 200 companies, has been long known as an effectively decentralised company. In his book, The Future of Capitalism, Paul Collier recalls the 1982 Tylenol crisis in Chicago, where 7 people died after drug tampering with J&J’s product. Contrary to traditional practice, “local branch managers [took] the initiative to remove all the Tylenol from supermarket shelves, promising the stores full recompense, even before senior management had time to react” (see my podcast with Collier about the book here).
Former CEO William Weldon, who worked his way up J&J from 1971-2012, spoke to the Wharton School in 2008 about “J&J’s decentralised structure”:
“The people who run our businesses around the world usually grew up in those markets, understand those markets, and develop themselves in those markets.
They can relate to the needs of the customer, whoever that customer may be.”
Committing the company to a liability of around $100m, that decision is considered an iconic lesson in crisis management forty years on. It is worth noting that this is a convex decision (they took an extreme approach, not a compromise) that was undertaken by a large group of people across multiple branches, illustrating how concave/convex decisions can be impossible to anticipate, whilst the perspectives of decentralised community members who “can relate to the needs of the customers” are able to make better, overarching market-facing decisions than centralised leaders.
At the same time, it was the central company that managed the legal service and designed a new product (better lids) to wrap up the crisis. This is a story, again, of decentralising the market-facing decisions and centralising those that were more operational. The challenge for DAO-builders is to construct institutions that make this balance work productively.
Modularisation is one lesson. Decentralisation for J&J, distributing autonomy to different sub-organisations, is different from web3 decentralisation, where effectively the executive committee at the top of J&J is turned into thousands of people. But copying J&J modularisation by creating modules within these DAOs can help define lines of authority.
Ukraine DAO is one example of this in practice, splitting its operations into many mini-DAO ‘pods’, from Fact Checking to Memes, where the core DAO just makes decisions about which pods to fund and to what extent. Designed in such a way that modules can be created and closed over time, it does not rely on the view that we can successfully design these institutions in advance. Worse is better.
“Leaders need to erect coordination totems that put everyone on the same page. The ultimate purpose of a coordination totem is to set the bounds for what ideas should achieve. It must be deliberately large in scope, but nonetheless clear.”
Johnson & Johnson’s coordination totem is their famous Credo, which Robert W. Johnson II, then chairman, set out in 1943:
“We believe that our first responsibility is to the people who use our products.”
By contrast, in the Uniswap forum, there was significant confusion about why the exchange was even funding education and political lobbying. When there are coordination totems to which most members are genuinely committed, the remaining blurred lines become less biting and more a space for collaboration. It helps members practise non-violent communication by putting everyone on the same page about the goals their organisation is trying to achieve and the sub-goals they are targeting in order to do so.
In summary, DAOs made up of individually influential voters are well-designed to pursue operational efficiency and make complex decisions, but if they attempt to bring in community governance (which expands what their organisation is able to do), they need to build new institutions to help this decision add stretch, over tension.
Primarily, they should do this by embedding institutional flexibility into the organisation. The capabilities to change as the organisation evolves and the relative benefits of who makes these decisions – community or technocracy – evolve with it. The will always be fundamental tension at play in this quadrant – after all, who decides who makes the decisions – but at the same time, there are good reasons to find oneself here. Building the same services as Uniswap and Compound but without any community involvement is worse than not decentralising at all. The above are principles to bear in mind as companies decentralise into this brave new world.