Protocol Thinking for Product People, Or How to Move Slow and Unlock Things

Protocol thinking is way more distinct from product thinking than I appreciated. Not different by degree. Not even different by kind. It’s different at the paradigmatic level. In short, it means:

  1. Expanding available moves over achieving specified outcomes
  2. Surrendering majority stakes in value capture and credit
  3. Higher standards and tighter reins around change governance
  4. Evolutionary search and selection over solutions engineering
  5. Care given via individual stewardship and collective goodwill, not exchange
  6. Measuring what’s given away to the wider world, not what’s captured
  7. Making constant failure and certain death generative
  8. De-centring humans amongst many other participating agents
  9. Designing a hard core that catalyses rare, outsized events for live players

First, traditional product thinking is outcome-based. One picks a target state, evaluates risks, and constructs archetypical cases to steer toward it. Success is arriving at the intended destination.

Protocol thinking optimises for something else entirely: an expansion of the moves available to players of the game. The protocol designer does have near-term goals, such as interoperability, legibility, and a low cost of compliance, but beyond them, they deliberately under-determine what gets built. That restraint is the point. Every outcome specified is a move taken off the board for someone else.

This is why indexing on outcomes, or even a cluster of them, fails. The upside a good protocol unlocks isn’t just hard to predict; it’s unnameable in advance. Steering toward a target state, however carefully chosen, narrows the space that’s meant to be widening. The question is whether others can now do surprising things in surprising ways neither they nor we could have specified beforehand.

Second, because protocol thinking solves for, essentially, unanticipated upside and future ecosystem optionality, one’s stance toward value capture and accumulated credit must change dramatically. Not softened. Largely relinquished.

Credit requires attribution, and attribution requires a nameable outcome. The moment one starts keeping score, one is pulled back toward the outcomes one can point at, which is to say back toward product thinking.

Value capture does the same damage more directly: every toll taken is a tax on the very moves the protocol was meant to make available. The aim is to expand others’ agency, not enlarge the designer’s share.

Protocol thinking in this sense asks for something product thinking never does: exercising one’s agency in setting the rules, then surrendering most claims to credit and value capture in exchange for others’ capacity to accrue them.

Third, inbound changes must be guarded far more judiciously. In product, a feature request is a signal to weigh against a roadmap. In a protocol, a proposed change is an amendment to the rules everyone plays by, and it deserves scrutiny proportional to that, especially when it foregrounds someone’s value capture.

This isn’t at odds with expanding available moves; it’s what makes it possible. The rules have to be conservative precisely so the play can be permissive. Move slow and unlock things.

Some open-source projects have closed themselves to outside contributions rather than drown in evaluating them. That’s one answer. The better one is explicit criteria and quality gates that make the bar legible before anything is proposed. The bar must be high because, in a protocol, the letter and the spirit have to be synonymous.

Product teams can lean on intent, norms, and a terms-of-service page. Protocols can’t. Anything the rules permit will eventually be done; anything extractable will be extracted, as maximal extractable value shows with grim reliability. A change that’s harmless in spirit but exploitable in letter is simply exploitable. If the spirit isn’t encoded in the letter, it doesn’t exist.

Fourth, it is more a game of evolution than engineering. Discovery-delivery loops, and even the market-level kinetics Wardley Maps describe, assume that search is expensive: one reasons one’s way to a good design because one can’t afford to try many. AI breaks that assumption.

With general intelligence on tap, generating and testing variants becomes cheap enough that search and selection outcompete design, especially in formalised domains. What survives isn’t what was best engineered, but what best satisfies global constraints, fits hyper-local last-mile requirements, and suits the tastes of the particular agents doing the selecting.

The protocol’s job is to be a good substrate for that search, not to anticipate its results. It’s akin to an immune system with its valence inverted: both generate variants no one specified and let selection sort them, but the immune system does so to eliminate the novel, the protocol to propagate it.

In a sense, protocol thinking is the Bitter Lesson incarnate for product people.

Fifth, because of points one to four, protocols don’t fit the constraints of market logic. Transactional exchange, the canonical quid pro quo, powers projects, platforms, products and services, but not protocols. Every funding mechanism one reaches for asks for something a protocol can’t give without ceasing to be one.

None of these work. That’s because each is a form of exchange, and protocols belong to a different paradigm: the gift.

Exchange settles accounts so both parties can walk away. A gift keeps moving and creates a loose, unenforceable obligation to pass something on. A protocol gives moves away and asks only that players keep the game going—ideally acting to make the game more interesting for others.

Gift economies are good at circulation but bad at upkeep. Nobody owes the giver anything, so nobody is on the hook for the roof. Protocol maintenance today rests on collective goodwill and on individuals who take up a role as benevolent dictators for a bounded period, carrying the cost of care and prophetic cognition for a while before handing it on.

The bound matters, because without it such stewardship hardens into ownership. Neither collective goodwill nor individual stewardship is really a funding model—at least not in the sense of capital compensating labour for technological development. How protocols pay for their own care, sadly, remains unsolved.

Sixth, in product thinking, most things one cares to measure are proxies for what the market legitimises as value. CAC, LTV, retention, conversion: each stands in for the same question. Namely: how much of the value generated does the generator get to keep?

Protocols, as established in point five, sit outside market logic. Their metrics can’t be proxies for capture, because capture is the very thing they forgo. Yet they still need something to be proxies for.

That something is value generated but not captured. It’s what a protocol is for, and it’s almost impossible to measure directly. No transaction happens. No one invoices. Nobody is counting. That’s the point. So, as with CAC and LTV, one does need proxies. Four will do:

Any one of these is cheap to fake. All four at once, sustained, is not.

They’re a set, not a menu. Game them for long enough and the four dimensions become indistinguishable from the real thing: a successful, sustainable protocol.

There are two things they deliberately don’t do, though. They don’t judge whether the play is good; that’s what the rules are for, per point three. And they don’t forecast. The four dimensions are organic, real-time signals, not targets one steers toward, because selection does the steering, per point four. The moment they become targets, for funders or anyone else, the paradigm of exchange dominates again and we bow out of protocol land.

Assuming the market can’t be the measure for a protocol, perhaps the M in TAM, SAM and SOM can stand for something else? Meaning added to the universe.

Seventh, product thinking treats failure as an event and death as a decision. Protocols get neither.

Complex systems always run partly broken, and “root cause” is mostly a social need for blame. Blame, like credit, needs attribution, which protocols have already given up. A protocol fails constantly, in small ways, and is held together by whoever’s there to keep catching it.

The danger comes when no one’s around. Disasters rarely come out of nowhere: the risks were known, maintenance was undervalued, and the system drifted. Death is a decision as much as an event, and most protocols have no one with the authority to make it. So they linger as zombies, held up by lock-in and inertia, with live rules and absent stewards. Anything the rules permit will eventually be done, so a dead steward and live rules is open season.

Managed death means planning the ending while the protocol is still healthy: sunset criteria, succession, deprecation timelines. It also means leaving something behind. Booms burn fortunes but leave tracks others can run on. A protocol that dies well does the same, leaving its spec, namespaces and habits as ground for what comes next. Move slow, unlock things, and when the time comes, die generously.

Eighth, in product thinking, not all agents are first-class. As a general rule, product thinking treats human agents as superior to machine ones. Even in contemporary situations, such as agent infra startups, human agents are, at worst, first amongst equals.

This primacy made sense when humans were probabilistic beasts and machines were deterministic engines. Humans decided what to do and how to say it, and machines did exactly as they were told. That narrowed the possibility space to four outcomes:

Every outcome traces back to a human. This was a consequence of computers, for better or worse, doing precisely what they’re instructed to do. It was also a consequence of humans, despite being ever more collectively corralled by sociocultural systems, creatively subverting instructions. The machine contributed nothing but fidelity, so the human was the only place judgement, and therefore error, could live.

Nowadays, a whole new class of machines isn’t so deterministic, and societal mechanisms of control are fracturing and fraying in step. Probabilistic machines reopen the possibility space with two outcomes the old one ruled out:

Once machines can subvert instructions too, outcomes no longer trace back to a single human. Judgement and error are distributed, and attribution, the thing credit and blame both depend on, starts to dissolve.

It gets more complicated, though, because the action is migrating to scales humans can’t inhabit.

At one end, high-frequency trading and agent swarms create miniaturised economies too fast and too small for humans to make sense of. At the other, exocapitalism is subliming capital from reality. Meanwhile, we’re having to work out Long Now problems we’ve been putting off for generations, problems too slow and too large for any one lifetime.

Lived experience has been here before. When steam replaced sail, the sailor’s hard-won feel for wind and weather gave way to stokers shovelling coal below deck. The ship still crossed the ocean and life aboard went on, but the people aboard no longer had a feel for how.

The upside is that all that complexity can be compressed into protocols. It just takes thought, human or otherwise. The cost is that human lived experience gets well and truly de-centred. It isn’t erased, though. It goes on, as rich as ever, but as one integral thread in a many-dimensional fabric of reality rather than the loom it’s woven on. For as long as we insist on being the centre, the fullest upside of protocols stays locked away.

Ninth, taken together, the above suggests protocols are better understood as event engines than as instruments for outcomes. Historians draw the same line between teleological time, where events are explained by where they were heading, and eventful time, where occurrences rearrange the very structures that made them possible. Product thinking lives in the first; protocols live in the second, hosting events that change what can happen next.

Take September’s Navier–Stokes result. Whatever one makes of the scramble over who got there first, the interesting part is what didn’t happen. No committee ruled on it, and the Clay Institute won’t even consider a proposed solution until two years after publication. Yet the work was taken seriously almost at once, because Lean had already checked it. Humans still had to confirm the right question was asked. The protocol acted as a hard core others could trust without rechecking, and that turned a result into an event they could build on immediately.

This holds whatever the shape of the distribution, but it matters most in fat tails. Where outcomes cluster around a mean, aiming at them works well enough; that’s product’s home turf, and its tools are built for it. Where rare, outsized events dominate, aiming is futile, because the event that matters can’t be named in advance, per point one. One can’t target the tail, only increase exposure to it. Event engines are how a protocol buys more draws for the live players ready to take them, and the upside is theirs to ride, per point two.

There’s a mirror image here too. Charles Perrow’s Normal Accidents showed that in systems both complex and tightly coupled, catastrophe isn’t an aberration but an expected property. Protocols invert the recipe: a hard core of tightly coupled rules, with loosely coupled play around it, which is point three seen from the other side. Keep the complexity, loosen the coupling where it’s safe, and surprise becomes normal upside rather than normal accident.


In parallel to the nine points above, I’ve been trying to pin down what protocols are like. Two methods help: likeness, through metaphors that show what protocols resemble, and difference, through the neighbouring paradigms they sit amongst and depart from.

Let’s start with likeness.

Gareth Morgan’s Images of Organization gave organisations eight canonical metaphors: machines, organisms, brains, cultures, political systems, psychic prisons, flux and transformation, and instruments of domination. Each one shows something the others miss. Reading Helen Czerski’s Blue Machine convinced me protocols deserve powerful metaphors of their own. My first candidates:

Taken together, the eight can be read through four broader lenses: time, space, structure and experience.

Signals and Games mediate how protocols are experienced, and Skeletons and Membranes inform how their adjacent reality is structured. Oceans and Commons determine the spaces protocols open up, which hold the option of becoming places once agents dwell in and care for them. Infrastructures and Genomes, in turn, steer how protocols persist and change through time.

These metaphors show what protocols resemble. The second method, difference, shows what sets them apart, by counter-positioning them against other paradigms for coordinating agents. Coincidentally, many of these begin with “p”, and the ones that don’t can be renamed so they do. So we can ask: what other Ps enable the coordination of agents?

I’ll adapt the Protocol Institute’s Protocols for Business glossary for a definition of our prime subject:

The other paradigms can be framed the same way. An X enables agent coordination through Y:

Read in order, they trace a coarse trajectory of civilisational development. They aren’t mutually exclusive, though. Several often operate at once, and usually one is first amongst equals, leaning on some of the others.

Mushed together, protocols are like oceans, skeletons and genomes, and unlike persons, prices and platforms. No one controls them; they’re strange rules unto themselves.


This all sounds, to be frank, radical, as if the old rules stopped applying sometime last Tuesday.

They haven’t. With some historical context, the shift reads less like a rupture and more like a change of season.

Consider World Machines. The guiding hypothesis is that civilisational systems run on roughly thousand-year lifecycles: about four centuries of emergence, four of mature dominance, then a rapid decline. They overlap rather than take turns. At any moment a Dawn, a Day and a Dusk machine coexist, and much of what passes for political, cultural and economic conflict is friction between them.

On that account, the Modernity Machine is in Dusk. It was a “pull” system, drawing everyone toward a single imagined future of progress and convergence. The Divergence Machine is in Day. It is a “push” system, producing radically different outcomes depending on local conditions, with no shared destination. And the Liveness Machine is in Dawn, its outlines only just visible.

Read against that, product thinking is a native of Modernity: pick the destination, then steer toward it. Protocol thinking is how coordination works when there’s no single destination to steer toward, and when the aim is keeping the game live for whoever is playing. Protocols are how Divergence holds together without converging. They will likely be how the Liveness Machine gets built.

None of this is new in kind. Whitehead observed over a century ago that civilisation advances by extending the number of important operations we can perform without thinking about them. Every protocol that works is one of those Whitehead advances: an abstraction that lets us ignore the machinery underneath. But the machinery is still somebody’s job. Sooner or later something made invisible turns out to matter again, and we have to choose to care for it. That is a Brandian return. It doesn’t reject the advance. It corrects course, recovers what was lost, and sets up the next jump. And it has to happen in a structured, protocol-like way, not as a retreat to the land. Advances without returns are how protocols end up as zombies.

What’s new is the timing. For over a decade, near-zero interest rates made it cheap to buy a market now and capture it later, so blitzscaling and platform lock-in were the rational play. With money no longer free, the logic of exchange gets dearer and the logic of the gift gets relatively cheaper. Add AI collapsing the cost of search, and the conditions favouring protocols are converging at once.

That doesn’t make the transition painless. Gramsci described the gap between an order that is dying and one that cannot yet be born as an interregnum full of morbid symptoms. The doom many of us feel is one of those symptoms, not a forecast. Crossing the Gramsci gap means adapting to the new machine’s logic rather than waiting for the old one to recover.

So the argument isn’t that other kinds of thinking lack value. It’s that a confluence of factors makes protocol thinking especially apt at this particular moment, and will keep it so for a long time.

The other paradigms keep their value. Let’s return to product thinking to see how.

The core of product thinking is that an agent has a job to be done. That endures. Much of human existence decomposes into jobs to be done, even though most people recognise that living is more than a bundle of tasks. What changes is where product thinking sits. Protocols don’t do jobs; they make jobs doable, by anyone, in ways no one specified. Product thinking lives on top of them, the way shipping lines live on top of the container. As with the other Ps, it’s rarely one or the other. It’s a question of which is first amongst equals.

There’s one more piece of context, and it’s about the field itself.

Protocol thinking is starting to harden into something more formal. The Protocol Institute’s Formal Protocol Theory group is doing the mathematical and logical modelling of protocols, building out the formal sciences underneath them, with applications from cryptography to distributed systems to healthcare. Its sessions read like a computer science curriculum being written in real time: process calculi and linear logic, stigmergy and swarm security, category theory, systems definitions formalised in Lean. There’s a Hilbert-style list of atomic protocol questions, and a textbook built bottom-up from homework problems. Even the working definition I borrowed earlier, a reliably repeatable behaviour, surfaced at its 2026 kickoff as the simpler way in.

That’s necessary but not sufficient. Computer science tells us what can be computed. It took software engineering, named at a NATO conference in 1968 (the same year the first Whole Earth Catalog appeared), to make computing reliable, budgeted and maintainable inside real organisations. Theory needed a discipline of practice before it could be put to work.

Protocol theory needs the same complement. My best guess for what to call it is decision engineering. If protocols mediate the sociotechnical infrastructure organisations run on, then what they solve for in applied settings is high-quality decisions at scale. Sometimes that means better decisions. Mostly it means making decisions self-evident, or unnecessary altogether.

The role already exists informally, as the protocol whisperer. What it lacks is legibility. A Decision Architect or Decision Engineer is a title organisations already know how to place, fund and evaluate. An architect owns a whole system across one or more domains; an engineer owns a narrow end-to-end slice of one. Like all titles, it’s a cover story. What it really buys is licence: the gonzo, high-agency aura to walk through otherwise solid walls and do unreasonable but allowable things.

Making decision engineering legible is the move product people are best placed to make. Product management was itself a translation layer, turning engineering possibility into organisational legibility. Decision engineering asks for the same translation, one paradigm along.


Ongoing gratitude to the Blygger contributors for inspiring the above.

Also of note: soon to be a book as part of the Protocol Institute’s 2026 PiBoWriMo.