The Throughput of Learning

Learning in the 21st century is not about acquiring more information, knowledge, or even insights. The goal is to maximize the throughput of invalidated assumptions. But you have to get there one step at a time.

When you first start learning, early in life, there is a bottleneck in the amount of information you have access to. You soak up everything like a sponge, because you are open and there is relatively little to absorb.

But very quickly, in elementary school, your access to information stops being the limiting factor. You take home a few giant textbooks, and suddenly the bottleneck moves to ways of structuring and contextualizing the information.

In high school, you learn a variety of methods to structure information — outlines, diagrams, underlining and highlighting, reports, essays, notebooks and binders. The bottleneck moves to your ability to synthesize this information, to turn it into new ideas.

In college, if you make it that far, the bottleneck moves to insight generation. You start questioning the world as given, and find that the juiciest intellectual rewards are ideas that shift how you view it. You start hunting for the revolutionary, the controversial, steering your learning toward the red pills of paradoxes and contradictions.


If you are lucky enough to go beyond this, the bottleneck moves once again: to your assumptions. They constrain your view, what you are allowed to see, and thereby the thoughts and actions available to you. You start getting a kick out of unearthing new assumptions, shining a light on blindspots that, by definition, you didn’t know you didn’t know about. This process is unbounded, because with enough examination, all your beliefs are revealed to be assumptions.

There are many ways to reveal assumptions. Interesting experiences, traveling, genuine conversation, and reading fiction all help you question your own point of view. But using throughput in particular to model learning gives us access to new frameworks and metaphors drawn from decades of manufacturing production experience. What does that history teach us about how to maximize throughput?

The first lesson is that quantity and quality are not opposing forces. You don’t have to sacrifice one to get more of the other. One can be used to enhance the other, like two sides of a coin.

There is a common misconception about throughput: that it is just a matter of “more widgets produced.” That it seeks to maximize total output, without regard to the quality of the thing being produced. But nothing could be further from the truth. In fact, throughput accounting is almost fanatical in its insistence that a sale has not been made until the final customer has purchased. This is in sharp contrast to mainstream cost accounting, which allows each producer in the supply chain to record a sale as soon as they’ve passed on the product to the next link in the chain. What this means is that the entire supply chain is held accountable to the most demanding standard of innovation and quality: the final customer spending her own money.

The most skilled practitioners of this yin and yang between quantity and quality is, of course, Toyota. Toward the end of his life, Taiichi Ohno, the creator of the Toyota Production System (TPS) that revolutionized manufacturing around the world, said a curious thing:

“I’m proud to be Japanese and I wanted my country to succeed. I believed my system was a way that could help us become a modern industrial nation. That is why I had no problem sharing it with other Japanese companies, even my biggest competitors. But I was very, very concerned that you Americans and Europeans would understand what we were doing, copy it, and defeat us in the marketplace. I did my best to prevent the visitors from fully grasping our overall approach. I explained it by talking about reduction of the seven wastes (muda)…and by talking about techniques with Japanese names like kanban.”

What Ohno’s quote and the most recent research seem to suggest is that we copied their system at the wrong level of abstraction. TPS is not a process for maximizing the throughput of finished products, as we’ve always assumed and admired. It is a process for maximizing the throughput of process improvements, even at the expense of short-term profitability.

Adding inventory between work centers, for example, would actually make the production process more adaptive in the short term, since interruptions in one place wouldn’t immediately disrupt the others. But the fragility of “single-piece flow” is its most important feature: by preventing extra resources from “flowing around” the problem, it forces everyone to improve the consistency and reliability of all parts of the process. And quality is improved as a byproduct. The system is designed to break in ways that surface the most useful lessons for improvement. The point of idealized goals like “zero waste” is not to obtain an outcome (which is literally impossible), but to create momentum in a direction that is likely to lead to lots of informative failures.

Similarly, modern learning is not a process for maximizing the throughput of insights, but for maximizing the throughput of learning process improvements. The best assumptions to invalidate in our quest for learning are assumptions about learning itself. This is why meditation retreats, globe-trotting, and having kids will always be net productivity gains, broadly defined: even a slight improvement in the machinery of learning (via a shift in perspective, for example) will pay dividends over time far greater than a mere few months of lost labor.


But here we’re presented with a dilemma: how does one learn about one’s own assumptions about learning? Since you don’t know what you don’t know, it’s difficult to learn anything about it. Our assumptions about learning seem to be permanently out of reach.

Another historical example gives us a possible clue: the history of inventory turnover.

Inventory turnover is a measure of the number of times inventory is sold or used in a time period. In the 1970s, U.S. manufacturing firms had an average turnover of 3.7 (in other words, they sold their full stock of inventory, on average, 3.7 times per year). The Japanese average was only a little higher, at 5.5. By the 1980’s, this had accelerated dramatically to nearly 20, with the highest-performing Western firms achieving turnover of between 30 and 80. Soon after, Japanese firms achieved an astounding three-digit inventory turnover, selling their stock 100 or more times per year. No one thought it could go any higher.

And then something weird happened: turnover sped up a little more, and suddenly turned negative. What happened was that production got so fast that the manufacturer was able to receive payment for the final product before they even had to pay their own suppliers. The customer could pay for and eat his hamburger before the restaurant has paid for the meat.

This shift, a linear change in speed that inverted the whole logic of business, presented us with some powerful, if challenging, questions: if we’re able to charge the customer before paying for the supplies, why should we be limited by our initial investment or working capital? Why not do the selling before production, so we know exactly what to make, and how much? What other stages of production don’t have to happen in sequence? We’re still working out the implications to this day: Kickstarter campaigns sell a vision to deliver on later; plentiful credit and falling barriers to entry make borrowing from the future easier than ever; the field of Lean UX provides us a ready-made toolkit of methods for validating before producing.

And note that this vast acceleration did not happen at the expense of quality. Over the same time period, we went from talking about defects in terms of “yield” (losses of more than 10%), to “parts per million,” an improvement of four orders of magnitude in under two decades.

This example illustrates one way blindspots can be revealed: by accelerating a system so much that its rules break, forcing everyone to confront its underlying assumptions. The shift to negative inventory turnover wasn’t technically or logistically difficult, but it required the reassessment of a deeply-held assumption: that the stages of production have to happen in a particular order.

This works because, when it comes to complex systems, faster is different. When you push a system to sufficient speed, it starts consuming itself, like a black hole turning inside out and emerging through the wormhole into an alternate dimension, where all the laws and rules are completely different. The system ceases to be a static backdrop where things take place, and becomes merely another unit of flow in a larger, more abstract system. We thought of “empty” space as an inert stage on which the planets roamed, until Einstein showed us that space is as much an actor as any planet. A few levels of abstraction later, and we’re starting to suspect that what was formerly empty space actually comprises the vast majority of mass in the universe.

The astronomical analogy points to something important: all the levels of abstraction are always present and possible, it’s just a matter of which level we want to operate at.

Take, for example, the phenomenon we observed with the evolution of learning, which is present in any complex, interdependent system: once you relieve a constraint, it moves. In manufacturing, relieving the constraint on production often moves it to distribution, as you try to get this new surplus of products out to stores. Relieving distribution moves the constraint to supply, as your suppliers have trouble keeping up. Relieving supply moves it to sales and marketing, since you’re now able to deliver as many products as you can sell. And so forth, forever.

What does this look like in a knowledge work organization? Let’s say you relieve the constraint on your software engineering team. This, of course, moves it to the product team, who has to develop product requirements faster in order to keep up. As the CEO, you could treat this movement of the constraint as a finish line, as evidence that you’ve succeeded. Pat yourself on the back, maybe take a break to do some “change management.” Or you could treat it as a starting line, just the table stakes in a different game at a higher level.

You could continue to accelerate this process, pushing your team to continuously identify and relieve bottlenecks as soon as they emerge. “Solving” for a particular constraint is no longer an objective in itself, but a way to keep the game going. Crucially, this doesn’t necessarily involve more work, just different work. Most problems you don’t bother trying to solve and can just ignore, because you know that pushing the system to the next level of emergence will change all the rules anyway, bringing fresh problems. You focus your resources only on the constraints keeping you from pushing the system just past the edge of its activation boundary, letting the momentum from constraint to constraint do most of the work.

This movement of constraints, jumping back and forth across your company, starts to take on its own dynamic. It has its own principles, its own habits, its own attractors. You can start to develop an instinct for where the constraint is likely to move next, how it will behave once it gets there, and what pattern its movement will follow over time. This instinct turns into a skill. Once acquired, it starts to be in your interest to keep the change coming, to keep everything just slightly unstable, teetering on the edge of coherence. If you can keep the system going without flying apart, the dance of the constraints starts to form its own picture, like the individual frames of a film reel accelerating and then suddenly merging together into a movie.

The difficulty in applying this concept to individual learning is that, in this case, you are the system. It’s a little disconcerting being accelerated, turned inside out, and then sucked into an alternate dimension where everything you were sure was true is wrong. Or worse, irrelevant.

The key is to realize that you are not a thing, which can be deformed and broken, but an environment. Like a factory, you could say. We have entryways and exits, windows and doors, well-lit areas and dark, forgotten corners. There’s a lot of machinery, much of it outdated, and it’s arguably more difficult to move around our mental machinery than the equipment on a shop floor.

It’s often said that knowledge work is totally non-linear, and thus completely incomparable to a production line. But there is another sense in which we do operate in a linear world: we operate in time. Time is even less forgiving than the tightest conveyor belt — there is no cord to stop it, and not even one minute that’s lost can ever be recovered. Not one hour can be reworked, and our supply is fundamentally limited.

But we work with ideas, which live forever, right? Maybe not. In another sense, ideas are highly perishable. The requirements for a piece of software take a lot of effort to collect, but quickly go out of date if not acted upon, as the needs of users evolve. This is fitting, as legend has it that Ohno was first inspired by the American grocery store and its on-demand replenishment of perishable goods.

But knowledge work is different! It doesn’t have the long setup times or changeovers of the old days. And yet it does: we know that the first stage of flow is struggle, as we get physically, mentally, and emotionally situated to the task at hand. And this doesn’t get any faster over time. We have to go through it every time we switch contexts.

It seems like what we’re left with as knowledge workers is a cavernous mental environment filled with heavy machinery, which we use to process highly perishable ideas through a rigidly linear flow of time, with long, delicate setups that have to be changed over every few hours or so. And no breaks — we’re always in there.

Our design problem, at least, is clear: we have to design our mental environment to maximize the throughput of invalidated assumptions, accelerating it to the point that the rules of our learning process break, thereby surfacing even more assumptions, which we can exploit to further improve this process.

Most theories of action sooner or later seem to settle on seeing as the most important leverage point in a situation, from Toyota’s “Go and see” to Boyd’s Observation/Orientation. I think what is required to make this model of learning work is, instead, a different way of listening. Specifically, listening for assumptions.

Listening for assumptions is a peculiar skill, not at all natural, that requires a continuous disassociation of what someone is saying, from the model they’re constructing in their mind as they go along. It’s like plucking fish from a stream, the actual words and sentences becoming just a delivery vehicle for potential entry points into their mental model.

But this is a lot harder than fishing, because there is no discrete activity you can focus on, to the exclusion of everything else. You have to actually listen to the words, even though they are mere delivery vehicles. You have to allow yourself to be emotionally impacted, because your intuition has a much higher bandwidth than your conscious thought. You have to care about what they care about, subordinate your interests to their framing, pretend their priorities matter so convincingly that you yourself are convinced. You have to stand IN the stream.

True listening requires giving up the prerogative of your own mental model. You have to allow them to set the rules of engagement, no matter how bizarre, so that they let their guard down and realize you are not a threat, because you have no intention of blaming them for anything. The way they set these rules will reveal their assumptions and constraints, which thoughts and actions are open to them. If you can tell an authentic story that speaks to these assumptions, you can break through, because stories speak to emotions expressed in the body, which fortunately refuses to go along with even our most well-reasoned rationalizations.

The price of such effective action is we have to be willing to give up the petty payoffs we cherish in our arguments with each other: not only the blaming but the cynicism, the martyrdom, the self-righteous indignation, the outrage, the winning, the making others lose, the being right, the making others wrong.

And after all that, you may still fail to convince them to your point of view. Are you willing to not win in order to keep the conversation going?

The interesting thing about constraints is that they are never on you. They are constraints on your context, shaping the space of possibilities you allow yourself to consider. You can’t change anyone’s mind (have you noticed?), but you may be able to change how they perceive their context. If you succeed, they may be able to step out of the criteria by which they enumerate their options. Life does not present itself in the form of multiple-choice questions. It is we who choose the choices, and we do it together.

What all this has to do with learning is that the deepest assumptions can only be revealed through experience and stories, not by reading books or having intellectual arguments. We do these things through the same old lens, and thus cannot examine the lens. It takes another free mind, reaching up and taking off our spectacles, to show us the cracks and the foggy areas. At some point, this way of listening turns into a way of thinking, as you apply it to your own thoughts. Unmoored from your own certain beliefs, you step back from what seemed just a moment ago to be your very identity, only to find that it is just a mental object. With each step backward, you distinguish your self one by one from bodily sensations, from emotions, from opinions, from thoughts, from principles, from values, from systems, from goals. They are all tools, to be taken up and put down again when no longer needed.

This backwards movement, if we are not afraid to embrace it, even accelerate it, starts to take on a pattern of its own once again. It starts to consume itself, emerging inward into a deeper, more complex flow. Moving backward with increasing speed, we start to feel as if we’re falling, the former selves flying by like the floors past a runaway elevator. There is no way to look down, to see where we’re going, only where we’ve been. But this provides just enough information to allow us to steer: toward discomfort, toward fear, toward our best guess of where the next bottleneck may lie.

Note from Tiago: this is, sadly, the end of my Ribbonfarm residency. Thank you for reading – it’s been a genuine pleasure writing for you. If you want to read more in the future, you can follow me on Twitter, sign up for my updates newsletter, or subscribe to my members-only (paid) blog Praxis

About Tiago Forte

Tiago Forte is the founder of training and consulting firm Forte Labs. His main interest is the intersection of design, technology, and modern work. Follow him at his blog or on Twitter


  1. “Point of view is worth 80 IQ points.”

    The article reminded me of *something*, and this turned out to be it:
    (so no surprises here).

    I think the tweetstorm has a few things that complement the article nicely, esp. regarding the motivation for all this:

    7/ What’s interesting is, each link in that causal chain is based on multiple
    assumptions, the failure of each one leading to solutions

    13/ Best solutions lie in category “didn’t know I didn’t know,” which means
    eliminating blind spots is more important than problem solving

    14/ Blindspots constrain your view of what is even a problem, not to mention
    what is possible as a solution

    27/ This explains why raw effort is so uncorrelated w/ success. W/ clear enough
    perception, u just always do the one highest-leverage thing

    28/ If the actions u take open the perception of others, relieving their
    bottleneck, that’s how u create massive compounding leverage

  2. Wow there are some quality gems in here. My favorite: “True listening requires giving up the prerogative of your own mental model.” Very true, very hard to do.

    Also, this: “The price of such effective action is we have to be willing to give up the petty payoffs we cherish in our arguments with each other: not only the blaming but the cynicism, the martyrdom, the self-righteous indignation, the outrage, the winning, the making others lose, the being right, the making others wrong.” I think historically this has been germane in every personal, social, and political context – and is ever so slightly more germane today in this current context of heightened sensibilities and capacities of righteous indignation.

    • Yes, its principles can be found everywhere. Every religion seems to have some form of it. It’s a “growing up” from Kegan’s stage 2 and stage 3 (communal thinking), to stage 4 (systems thinking). Your own personal payoffs are subordinated to the goals and responsibilities of your current role, which is to teach, persuade, and enroll.

  3. Boyd’s insights included detecting invalid assumptions. In The Essence of Winning and Losing (1995), he described how to accomplish this with a “many-sided, implicit cross-referencing process of projection, empathy, correlation, and rejection (across these many different domains or channels of information).” Boyd preferred the phrase ‘mismatch detection.’ For the process to remain relevant, Boyd noted “an evolving, open-ended, far-from-equilibrium process.”

  4. Hi Tiago,
    Great read. Subtle theories on – phases of learning(collection of knowledge, organisation of knowledge, generation of ideas, generation of insights, identifying & invalidating assumptions), existence of abstraction(with astronomical analogy) , and shifting of constraints were articulated really well.

    Just few questions popped up-
    * Are we just meant to chase the bottlenecks at each phase, where would this take us ?
    * Why can’t we jump bottlenecks non-linearly, than moving linearly ? using a mentor or understanding about a system and its projections than discovering via first hand experiences etc.

    Looking forward to your thoughts.

    • Thanks!

      1) Yes, there is no end to the shifting of bottlenecks. The mountain has no top. We know this because a company without bottlenecks would be able to serve infinite customers at infinite speed, thus causing the universe to implode. Reality is mostly made up of constraints, not mass

      2) If I understand correctly, that is what I’m proposing in the second half of the essay. That instead of treating each bottleneck as an individual project, one at a time, we accelerate and create a constant flow of bottlenecks. It’s like graphing an accelerating object: The speed graph has an exponential curved line, but the acceleration graph has a straight line to describe the same movement. The latter is describing the same thing at a different level of abstraction

      • Thanks for the response Tiago.

        On point 1) If there is no end at all why should we chase them anyway. Why can’t we just stay within the existing constraint without trying to shift them. I presume buddhist monks and spiritual saints try to stay in this state of mind rather than fighting against these fleeting phases of life – material, emotional, physical, mental, intellectual etc. Am I missing anything ?

        • Tiago Forte says:

          No you’re not missing anything. But you’ve reached the rock bottom of the abstraction ladder, and from here only philosophy can help you. Why improve? Why strive? Why live?

  5. Kai Balder says:

    Thanks Tiago, very thought-provoking read! One of the best articles I’ve read on this topic!

    I like your bottleneck approach a lot, one of the things that helps me organize my learning is the realization that the fundamental bottleneck is the limited capacity of our brain – which I think is also driving the bottlenecks to learning you’re describing

    The “non-genetic” evolution of our society (predominantly through knowledge generation) has happened at a much much higher pace than genetic evolution. As a result we ended up with a caveman brain in a highly complex world – every day our brain becomes less prepared for the world we live in

    In order to operate in this complex world, we need to use our limited brain resources as effectively as possible:
    1) we do not have the brain storage capacity to store all raw data thrown at us (after early life), so need to start using the relational capacity of the brain to store it in a more efficient way and use external storage capacity (e.g. OneNote, Evernote, etc.)
    2) we do not have the brain capacity to have more than a couple of variables in the forefront of our consciousness and process them (e.g. try multiplying 18 * 78 without using a piece of paper) so we need to externalize the variables and the connections between them. And we have to investigate them one by one (i.e. break them up) to have them fit into our limited forefront of consciousness

    Please correct me if I’m wrong but I have the impression we use the above two features continuously to overcome the different bottlenecks you describe (first structuring, then synthesis, then new insight generation, then invalidating assumptions)

    • I would say this is the conventional opinion about bottlenecks in our learning ability. Now, can you go through and identify the underlying assumptions behind the hypothesis you laid out? If you can do so, then find the weakest of the assumptions, then find a situation in which that assumption doesn’t hold, you’ll be on your way to the next level of abstraction. I guess that is the takeaway homework for this post :P

Leave a Comment