You have to choose between speed and safety. You can't get both. That's what common sense tells us, and common practice confirms.

And then, there are the roundabouts. So simple and yet they miraculously increase both throughput and safety. Roundabouts reduce delays by 89%[1] and cut fatal crashes by 65%[2].

To see why this is the case, let's compare it with the other two ways to regulate intersections: manual traffic control and traffic light control.

This essay is in Autonomy and Cohesion series but is self-contained and it doesn’t require reading the previous essays.

Manual, Lights, Roundabout

Manual traffic control is a centralized form of coordination, where the behavior of the affected agents is directed by a single person. It relies on officers’ senses (extended with radio reports) and judgment.

Traffic lights are an automated mode of control, with varying levels of adaptability to traffic conditions, ranging from none (when using only timers) to the use of simple sensors, and ultimately to smart lights utilizing IoT and AI. They have better throughput than manual traffic control, but would not make a specific decision based on direct perception and judgment of the situation.

In roundabouts, there is neither a person nor an algorithm. The physical properties, together with some simple rules, serve as constraints that produce self-organized coordination behavior. The emerging order results from each driver following basic rules (yielding to circulating traffic) and relying on personal judgment within the space between these rules. It is a matter of distributed cognition made possible by a protocol.

A prominent difference between manual and traffic light controls on one side, and roundabouts on the other, is the mediation of the coordination. For a driver, Alice, to coordinate with another driver, Bob, there must be an intermediary, either a human (a police officer) or a machine (a traffic light system). With a roundabout, coordination is achieved without an intermediary.

In all three situations, a protocol layer is present: thin in manual control, thicker with traffic lights, and fully dominant with roundabouts. In manual traffic, the embedded protocol consists of traffic laws, road markings, and standardized hand signal conventions that drivers must understand. Because a police officer can override or bend any of these cues on the spot, this protocol is very thin. Traffic lights codify rules such as "if red, stop" and "if green, go." There are also internal, hidden rules that can be inferred from the simple case of using only timers. In the technology-heavy case of smart lights, these rules remain opaque. The roundabout directs through physical affordances, rules communicated by specific signs, and the general roundabout rule "yield to circulating traffic."

Manual control is more adaptable to changing traffic conditions than even the smart lights. But this adaptation is a matter of reacting to congestions and emergencies. Roundabouts, in contrast, self-adapt continuously: drivers naturally adjust speeds and gaps, which smooths the flow.

Manual control is fragile: if the officer is distracted or absent, control collapses. Traffic lights are reliable, but a power outage or malfunction can halt the system. Roundabouts are extremely robust: drivers will instinctively slow and negotiate gaps even without signage.

Manual control scales poorly: one officer can effectively manage only one intersection or one direction. Traffic lights scale better – one city can network many signalized intersections, though each requires its own equipment. Roundabouts have high scalability: no extra technology is needed to add traffic (beyond roadway design), and even busy multi-lane roundabouts operate without controllers.

Manual control incurs ongoing labor costs and is expensive over time. Traffic lights require capital for installation (including poles, wiring, and sensors) and periodic maintenance. Roundabouts have high up-front construction costs (landscaping, redesign) but minimal operating costs thereafter.

These three ways of coordinating can be seen as exemplars of the three main coordination modes: leader, platform, and protocol. But, unlike traffic regulation, where there is a data-supported general agreement that the roundabouts are the most efficient and effective way, that is far from being the case in other situations of coordination and governance.

Link & Think is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Leader, Platform, Protocol

We need ways to deal with each other, because we share commons, and because the social structures we build tend to create even more coordination problems.

Initially, coordination was a matter of ritual and hierarchy. The invention of writing enabled record-keeping, planning, taxation, and administrative protocols. Armies were coordinated through clear chains of command. Later, guilds and trade leagues developed systems of rules and standards to regulate trade. The industrial era brought standardized workflows, followed by cybernetics and operations research, which introduced feedback control and mathematical optimization. Today, we have internet protocols and platform algorithms.

In all cases, the coordination is achieved via some combination of leader, platform and protocol, with one of these modes dominating.

Leaders

Leaders are highly valued in business. The top 10 CEOs’ salaries are in the range between 80 and 200 million dollars. Leaders are valued even higher in politics. The US election campaign consumes billions of dollars.

We want to be saved by special people. Some want to be saved by a special person holding a sword; others want to be saved by a special person holding a pen.

Our narratives tend to focus exclusively on the superficial deportments of Special People — the retreating self-effacement of the sophisticated Great Bureaucrat vs. the charismatic theatrics of the populist Great Man. But the shared epistemic commitment to Greatness as an explanatory model goes largely unchallenged. We are so busy fighting over the relative merits of Pareto Lions and Pareto Foxes, and accusing each other of elitism, we forget to question the shared premises of all elitism.[3]

The common desire to be saved by special people is inspired by fairy tales, fantasy stories and action movies.

We see leaders as desirable or inevitable in every situation involving coordination and governance. If we reuse the source of inspiration — fairy tales, fantasy stories and action movies — also as a source for classification, we can plot leaders in dimensions from evil to good, and from stupid to smart:

          | Evil       | Good
----------|------------|--------
Stupid    | Harmful    | Useless
Smart     | Calamitous | Precarious

The outcome is either bad or unreliable.

A historical example of stupid and evil is Idi Amin; for a contemporary, a random pick from the MAGA elites has a high probability of being spot-on.[4] Moving to the right, Neville Chamberlain seems an excellent fit. He had good will, but — putting it mildly — lacked strategic foresight. Down-left, that would easily be Stalin. Moving to the right again, a good pick from the same lands would be Mikhail Gorbachev. A more contemporary example is Jacinda Ardern. And since that's the only hopeful quadrant, a third example, this time not for a country leader but a company (and platform) leader, would be Jack Dorsey.

Leaders in the lower right quadrant don't last long. They face personal, political or structural pressures. Gorbachev's reforms unleashed forces that he couldn't control. After an impressive crisis leadership, Jacinda Ardern abruptly vacated office mid-term. Jack Dorsey made hundreds of millions believe that social media platforms can be a good thing, then the board replaced him with Parag Agrawal, and finally Musk’s takeover in October 2022 ousted Agrawal and, by extension, any remnant of Dorsey’s approach.[5]

Leaders in the first column of the table stay in power by relying on a small group of essential supporters whom they reward to maintain loyalty. Leaders in the lower right quadrant depend on the support of a larger group of people, making it harder to satisfy everyone, which leads to shorter tenures.[6]

Leaders in the lower-right are not precarious only for being replaced by leaders from the other quadrants. They may not be replaced at all but simply drift to another quadrant.

Most leaders use their power to stay in power and need platforms to amplify that power.

Platforms

Platforms are commonly used as a shortcut for digital platforms. And that means a client-server architecture and business model, providing centralized services. If we look at platforms as a more general coordination paradigm, then we need a more inclusive definition, like this one:

A (socio)technical intermediary that coordinates and governs interactions among participants.

Classes of platforms are: banks, representative democracies, nation-states, churches, and, of course, the digital platforms.

Examples of digital platforms are Facebook, X-twitter and LinkedIn in social media, Uber, Bolt and Lyft in transport, Airbnb and Booking.com in accommodation, Amazon and e-Bay in e-commerce, Stripe, PayPal, Revolut, and Wise for payments. Many digital platforms disrupted traditional businesses in a software-eating-the-world fashion, demonstrating how a transport company can be profitable without owning a single vehicle and how an accommodation company can be profitable without owning a single hotel.

One distinction between digital and other platforms is that the former take advantage of network effects. Facebook attracted the same number of participants as the Catholic Church, only 200 times faster.

Digital platforms have a small power locus, either a single authority or a small group. That can be a sole owner like Musk of X-twitter, or a founder having control over the company through the majority of the voting rights, which is the case of Mark Zuckerberg for Meta (Facebook), and Zhang Yiming for ByteDance(TikTok). Similar is the case for Airbnb, Spotify and Stripe, with the difference that the majority of the voting rights are collectively owned by the co-founders.

Platforms provide intermediation, structure and control.

Platforms provide intermediation by connecting participants and enabling different kinds of collaboration. Commercial banks are intermediaries between borrowers and savers. Central banks are intermediaries between commercial banks, governments, and the broader economy. In representative democracies, elected representatives act as intermediaries for the distribution of the commons between citizens and for translating public needs into policies. A nation-state is a platform where the government acts as an intermediary between citizens and national affairs, enforcing laws, managing public services, and regulating interactions within its territory. Churches serve as intermediaries between believers and God, as well as between believers themselves. Digital platforms connect users for social interactions (Facebook, X-twitter) or service providers with consumers (Uber).

Platforms provide infrastructure and affordances. Commercial banks deploy branch networks, ATMs, and digital structures for online services to enable financial transactions across. Central banks operate payment and settlement systems infrastructures for interbank clearing and monetary control. Representative democracies maintain legislative frameworks, electoral systems, and bureaucratic processes that structure policy-making. Nation-states build public works to enable service delivery and civic participation. Churches build sacred spaces — cathedrals, chapels, and altars — designed to channel worship through architectural cues, liturgical furnishings, and ritual objects that constrain and enable congregational behaviors. Digital platforms offer computing infrastructure that shapes how users navigate, create, and consume content or services.

Platforms provide control and governance through a system of rules and enforcement mechanisms. Commercial banks apply governance, risk, and compliance frameworks. Central banks provide payment and settlement rules, collateral eligibility rules, and liquidity mechanisms that coordinate funds transfers among banks. Representative democracies provide the rules of electoral engagement and legislative procedures, defining who may stand for election, how votes are cast and counted, and the formal debate and amendment processes in parliaments and committees. Nation-states enforce specific legislation within their geographical boundaries. Churches govern through canon law and ecclesiastical courts. Besides that, there are specific sets of rules like those of Saint Benedict. It was the first systematic codification of monastic life. Digital platforms govern users via Terms of Service, enforced by automated algorithms and human moderators.

Protocols

Protocols are best understood through the famous quote of A. N. Whitehead:

Civilization advances by extending the number of important operations which we can perform without thinking about them

It also shows why protocols are rarely thought of as a solution to complex problems. They get visible only when they are dysfunctional.

Protocols are not just difficult to see, but also difficult to define. Yet, here are a few attempts:[7]

A set of explicit or implicit rules that guide behavior and facilitate coordinated action.

A set of rules, norms, incentives, and design affordances that work together to translate a potentially chaotic range of behavior into a more orderly result.

A protocol is an engineered argument.

Protocols are coordinated through rules, but those rules don't come from a ruler or an intermediary that a select few can control. If the protocols are good, then their rules apply equally to everyone.

Unlike traditional technologies created by private corporations, governments, or even open-source communities, where an individual auteur architect or engineer might powerfully shape major tradeoffs by fiat and with finality, protocols tend to evolve through, and as, structured arguments that no one actor can dominate.[8]

Protocols provide predictability through their hardness, much like hard metals (both in terms of being difficult to change and difficult to obtain) did for money and stones for buildings, and the way institutions do for rights and obligations.

Protocols often spread slowly (example: handwashing), but once established, they may last long.

Good protocols are cautonomous. That means they enable cohesion with minimal or no loss of autonomy. Cohesion here means the ability to form a whole or work as a whole. In all cases, cohesion is achieved through constraints, resulting in reduced autonomy for participants. Good protocols may decrease some part of the autonomy — participants are constrained by the rules of the protocol to participate — but the overall autonomy of each participant is increased. How this might be the case, I have written here and from the perspective of the uselessness of the decentralization/centralization debate, here.

Single Point of Failure and Power

There are several dimensions on which leaders and platforms cluster close to each other on one side, while protocols go to the other. The first one is about mediation.

All coordination is mediated.[9]The mediation layer can be linguistic, technological, or cultural, and most often a combination of these. But there is a difference when the mediator has agency or is controlled by a single agent or a small group of agents. While there is an intermediary in the cases of leader and platform, in the case of protocols, there is none. That doesn't mean the problem of trust is dissolved. If traditional transactions require both parties to trust a bank (the source of hardness), they need to shift their trust from an institution to a technology in the case of using crypto protocols.

An intermediary doing coordination makes it both the weakest and the strongest point in the network. It is the weakest because it represents the single point of failure (SPOF). If that is what makes coordination possible, the coordination breaks if the intermediary fails (however you decide to define failure). At the same time that dependency concentrates power, which is also used to increase the dependency, the lock-in.

In a similar way to energy in physics, in socio-technical systems, there seems to be a law of agency conservation.[10] A controller with high agency means less agency for the average participant. Then, coordination with non-agent, a good protocol, is what can bring high average agency and reduce power asymmetries.

Digital platforms need special attention. They are not just intermediaries. Digital platforms impose a specific economic model that reduces the natural self-organization of interacting agents. Amazon provides a market where the market forces are manipulated by Amazon algorithms. Buyers and sellers are not negotiating parties in a free market economy but serfs in a feudal land. Facebook ostensibly provides a town-square, which is not even a metaphorical town-square but a market and a market which is not a real market since the agency of the participants is reduced by the Facebook algorithms.

Reform or Replace?

Should we strive to get better leaders, knowing that good leaders don’t last long, or find a way of organizing without leaders? Should we reform our platforms or replace them? Or maybe remix?

There are self-organized, leaderless organizations, but only on a smaller scale. Or, if there is a bigger scale, it applies to specific domains such as online collaboration, but not to a large production facility or a polity. Can there be a bigger scale leaderless coordination? Or is there a need for aggregation the way that's traditionally done, or can it be achieved through the coordination of smaller-scale organizations?

And what about platforms?

In his recent book Against Platforms, Mike Pepi criticizes digital platforms but defends institutions. He doesn't treat institutions as platforms.

I agree that not all institutions are platforms, or rather that some have a more important role in society in addition to their intermediary role. Hospitals, universities, and museums belong to that category of institutions. But other institutions, like notaries and banks, certainly are platforms.

So, reform or replace? When I asked this on Substack, context-free, this is the reply I got:

The Catholic Church is a large-scale platform engaging many Christians, but there were always alternatives that we can rightfully call protocol-based religions. First to oppose the need for a church among Christians were Paulicians in the 7th century in Armenia, followed by the Bogomils in the 10th century in the Balkans, promoting a simpler, more personal form of Christianity. Then, the Cathars formed in the 12th century, and a bit later, the Waldensians. Both appeared in France with some spread in Spain and Italy. By the end of the century, in the Low Countries and Germany, another Christian movement appeared which rejected the need for institutions to practice faith, the Brethren of the Free Spirit. Following them were the Anabaptists in the 16th century and the Quakers in the 17th century in England.

There is a lot to say about these movements, but for our purposes, two notes will suffice. First, all of these were protocol-based, platform-less Christian movements, but none of them had as large a following as the institution-based Catholic, Orthodox, Anglican, and Protestant churches. And the second is that only the Brethren of the Free Spirit remained leaderless.[11]

In business, there is this practice, Management By Exception, where employees work more independently and only involve their managers when there are exceptions to normal operations. Can this be re-imagined as Managers by Exception? Or more generally, leaders by exception. After all, manual traffic control was common before, and now it is only used in exceptional cases. There are plenty of such examples in the governance of tribes. In the harsh Arctic winters, Unuits would aggregate into large groups for cooperative hunting and resource sharing under the guidance of respected elders or skilled hunters, but such leadership was not needed and not practiced during the milder summers. The situation was similar for the Nambikwara, who lived in two different social orders during the rainy and dry seasons. Greaber and Wengrow explain:

Chiefs made or lost their reputations by acting as heroic leaders during the ‘nomadic adventures’ of the dry season, during which times they typically gave orders, resolved crises and behaved in what would at any other time be considered an unacceptably authoritarian manner; in the wet season, a time of much greater ease and abundance, they relied on those reputations to attract followers to settle around them in villages, where they employed only gentle persuasion and led by example to guide their followers in the construction of houses and tending of gardens. In doing so they cared for the sick and needy, mediated disputes and never imposed anything on anyone.[12]

Again, these are all small-scale examples in non-industrial settings. But if they can't be transferred, they can at least inspire some social experimentation.

Back to platforms, some partial replacement by protocols is already taking place. Bitcoin and Ethereum bypass both central and commercial banks. ActivityPub and ATproto moved tens of millions of users away from X. IPFS offers host-independent file storage and sharing.

And then, of course, there are more roundabouts replacing traffic light intersections. But while roundabouts are good when they coordinate vehicles, they are not so good, on their own, when the participants include cyclists and pedestrians. We need a better protocol to handle this case.

If there is more experimentation and innovation in protocols, they can replace more coordination intermediaries for a bigger part of the population. The three coordination modes will continue to coexist, but hopefully in different proportions.

Better protocols can initially increase the cognitive load of the participants. They can be experienced as an inconvenience or create confusion. A better protocol may resemble the Magic Roundabout in Swindon, UK, compared to a traditional one.

It has five mini-circles around a sixth bigger one. This creates a configuration of seven roundabouts: an outer circle and the five mini-roundabouts going clockwise, and the inner circle going anticlockwise (way more complicated than the sign shows). It sounds like a nightmare, and many drivers believe it is. And yet, it has been there since 1972, and statistics indicate it is both faster and safer to go through compared to similar junctions.

A new protocol can feel difficult at first, then transition into a managed phase, until it reaches a point where it is followed instinctively — a real advancement, in Whitehead's sense.

First published on Link & Think.

[1] Retting, R.A. & Mandavilli, S. & McCartt, A.T. & Russell, Eugene. (2006). Roundabouts, traffic flow and public opinion. Traffic Engineering and Control. 47. 268-272. https://www.researchgate.net/publication/286979578_Roundabouts_traffic_flow_and_public_opinion

[2] Gkyrtis, K., & Kokkalis, A. (2024). An Overview of the Efficiency of Roundabouts: Design Aspects and Contribution toward Safer Vehicle Movement. Vehicles, 6(1), Article 1 https://doi.org/10.3390/vehicles6010019

[3] Strange New Rules, Protocolized

[4] See Incompetent or Evil: A False Dichotomy by Paul Krugman

[5] Pekka Kallioniemi provides a good summary of what Musk turned X-twitter into in this thread.

[6] The Dictator’s Handbook. (2017). https://www.hachettebookgroup.com/titles/bruce-bueno-de-mesquita/the-dictators-handbook/9781610390453/?lens=publicaffairs

[7] All are taken from products of the Summer of Protocols program.

[9] Protocol Reader

[10] That’s what Vygotsky found out back in the 1920s, not just for coordination but for any action.

[11] Early Quakers were also leaderless, but that changed in the 18th century.

[12] Graeber, D., & Wengrow, D. (2021). The Dawn of Everything: A New History of Humanity. Farrar, Straus and Giroux. https://us.macmillan.com/books/9780374157357/thedawnofeverything/