Exactly enough text

A personal digression before I continue to riff on the idea that corporate knowledge is largely illegible to text. I’ve been deeply invested in Wikipedia for close to 20 years, enough for me to log over 60,000 edits. The internal machinery of the place fascinates me, especially the administrative backrooms that resolve and sometimes foment disputes. It is altogether unique on the modern web: a cross between encyclopedia-as-MMORPG and non-hierarchical free-culture experiment where it’s okay to adopt typos, or add specific snippets of text to the parts of articles where readers are systematically least likely to venture, or run around posting {{R with possibilities}} on hundreds of pages. Nobody is assigned to onboard you on your first day, nobody will hand you a learning package,Although you might be welcomed with a plate of cookies (opens in new tab) anyway. and there are no org charts or 1:1s or formal reporting structures. And yet somehow it all works!

Now I don’t mean to claim that the self-organized Wikipedia community is especially well-functioning. There’s a metric ton of cruft and spam and undisclosed paid editing (opens in new tab) and slopwriting; these are massive community time-sinks that threaten the quality and integrity of the encyclopedia,Footnote 1See e.g. this WT:CSD discussion (opens in new tab) (archived Jul 2026). and you can argue defensibly that the community is ill-equipped to overcome them. There are also real governance problems worth their own posts, as well as frequent conflicts with the Wikimedia Foundation over technical priorities, budgeting, and even hiring-and-firing decisions.Footnote 2See COMMTECHGATE (opens in new tab) (archived Jun 2026). And yes, the community sometimes — wrongly in my view — fails to sanction editors for uncivil or unproductive behaviour on the strength of their past contributions, as though the “brilliant jerk” problem were not worth solving. But Wikipedia persists amid these threats and fears, and one finds it hard to imagine a world where the latest celebrity death isn’t painstakingly documented within 30 seconds of the announcement.

There are exceptions to most rules, and they’re especially productive when the rule is copied from a top-10 website like Wikipedia. So here’s a story about a wiki that didn’t work so well. A couple weeks after I’d started my first job, I was told that the division would be moving its corporate knowledge to a MediaWiki-based wiki. Perfect, I thought, a great opportunity to help build something technical where I had some domain knowledge. So I started boldly editing the wiki, as one accustomed to enwiki’s principles does,Footnote 3Wikipedia: Be bold (opens in new tab) (archived Jul 2026). It did not occur to me that I was the only person contributing without having been asked to, enwiki guidance notwithstanding. by building a navbox and networking the in-progress articles via wikilinks. More content started to appear, largely contributed by a couple editors who were evidently working through a prearranged set of topics given the speed with which they created pages. I didn’t know the source of the content or where the list was coming from, but I figured it would be addressed at the meeting that I’d just been invited to on the topic.

The meeting didn’t clear up much for me; nor did the second one held a week later. There was no third meeting, and around that time I saw the trickle of edits in the Recent Changes feed peter out entirely. Only later did I notice that the project had been both conceived and wound down around October. This is incidentally when all staff have mid-year performance discussions with their supervisors.

On one level, stopping the project cost us nothing, as the wiki had attracted only a few contributors and had managed only to aggregate pre-existing content.If you already know chapter and verse of what you intend to document, either you’re an expert or you’re writing a tutorial. There’s nothing wrong with tutorials, but they’re only one of the four key types of documentation, and you should know that going in! On the technical merits it was probably too surface-level to enable a newcomer to craft an internal-standard-compliant WHERE clause or whatever, so it wasn’t hard to write the whole thing off. But on another level, the slow death of the wiki extended a real problem for us, as we still didn’t have truly useful technical documentation, which made my own onboarding significantly more challenging. Worse yet, it dampened any managerial impetus to tackle shared-understanding debt for years.”We already have a solution for this, check the wiki! But can you ask Alice for the link? I don’t have it bookmarked.”

Everyone has a story about a failed documentation project, which makes mine boring but also universal. It’s another example of the pilot-contronym problem I wrote about earlier: the wiki became a not-wiki the minute incentives other than knowledge-sharing were attached. But it’s also a great object lesson for how text can be drafted for reasons orthogonal to communication, or if you prefer, how it’s hard to make text reliably fix meaning and purpose at once. There’s always some additional piece of critical information floating undocumented, below the text, that would reframe your understanding if revealed.

My failure to notice this meant that I contributed too much to the wiki. My categories and navboxes and my network graph of articles that an Obsidian user would have envied — these ultimately served no purpose except to disappoint me when they didn’t stick. But the rest of the group wrote exactly enough text to achieve the intended output. This is what Beer meant when he observed that the purpose of a system is what it does (opens in new tab); the purpose of our wiki was to be built, not to become an interconnected Wikipedia-style resource that smuggled enwiki cultural relics into our workplace. Yes, you can argue that the knowledge transfer routine has been reified, but the point isn’t whose point of view was correct, but rather that there were multiple points of view in the first place. This is how Joe Litobarski sees it, writing the following in a refreshing post on POSIWID:Footnote 4Joe Litobarski, POSIWID (opens in new tab) (archived Apr 2026).

the system – and its purpose – always emerge in relation to a point of observation, a perspective… when managing a complex system, it’s not just helpful to observe that system from different perspectives – it’s necessary, because each perspective actively helps constitute what the system is. The boundaries, the behaviour, even the purpose of the system shift depending on the position, experience and assumptions of the observer. Any model of the system that wants to regulate that system should account for multiple perspectives.

So you can conceptualize the purpose of the wiki as, depending on the observer, a densely interconnected information hub, a device to achieve the October incentive, or even a PowerPoint slide with a screenshot. But there’s another interesting distinction between the two wikis. Wikipedia is text-only by nature;Modulo the recent rise of Discord, which has actually become a pretty significant hallway for ArbCom governance purposes. you can’t walk down the hall, bump into another editor, and trade ideas about a problem she’s working on, because you usually don’t even know who any of these people are! This leaves you with no choice but to document fully and clearly, for fear of your ideas and editorial suggestions becoming a private language (opens in new tab) of sorts and failing to gain any traction. My team’s wiki was the opposite: we all tended to communicate orally, so for Conway’s law (opens in new tab) reasons the wiki became something you needed talking to understand. Oral-first may not be a particularly efficient way of doing some things, like onboarding newcomers or training juniors or getting to a consensus on a complicated technical question, but it’s how the team was structured for other reasons. As a result, even setting October-style incentives aside, it’s highly unlikely that a prodigious amount of writing could have moved the needle on knowledge sharing. Volume was never what the org was selecting for. And in a funny reversal of the Wikipedia case, this turned our wiki into a private language too. When you have to ask Alice for the link, you’re just admitting that the real knowledge lives outside the text.

Options analysis

So how much content should I have written for the wiki? How many edits should I have made to Wikipedia? How long of a recommendations document should anyone write for their next strategy meeting amid the fundamental illegibility of knowledge? It’s not easy to know when to stop writing unless you have a clear picture of the incentives at play, which arise from and are consistent with what Nils Brunsson (opens in new tab) calls ideology: the set of “ideas and values prevailing among organization members about the organization and its environment”.Footnote 5Nils Brunsson, The Organization of Hypocrisy, 1991, p. 16. These values are probably more abstract than mere if–then rules, which don’t foresee the sheer breadth of possibility in highly volatile or uncertain contexts.Shades of the civil-versus-common-law distinction I discussed in “The optimal amount of framework wrongness”. But abstract doesn’t mean vague; on the contrary, Brunsson argues that action-oriented organizations use ideology to render decisions largely unnecessary. If the decision’s been de facto made in advance through a common standard, everyone has a pretty good idea what the solution to a novel problem ought to be. But of course this requires that the org have adopted a certain higher-level clarity of purpose as a precondition. This isn’t quite the same as Hayek’s tacit knowledge, although they look similar if you squint. Hayek cares about the fundamental impossibility of accessing the working knowledge that accretes within the pockets of your org. Brunsson’s conception of ideology may have it supplant written decisions, but the ideology can always be written elsewhere, or at some earlier stage. But what unites the two is how they both make it impossible for the outsider to reconstruct the decision-making process.

And when it comes to reconstruction, there is no better example of the interplay between Brunsson’s views and writing than the options analysis. It’s a basic tool that you’ve seen dozens of times, because it’s fundamentally a rational way of making decisions. You think you’d be hard-pressed to find a substantive objection to it, but Brunsson comes up with a good one, noting that options analyses leave significant room for uncertainty and doubt. If every possibility is alive with subtle advantages and hard-to-anticipate second-order harms, then the analysis has a paralytic effect on the org, for which the myriad downsides to any action are now eminently legible. The firm emerges less confident about the right path than before the analysis had been written. This all makes it unsurprising that in The Organization of Hypocrisy, Brunsson says that action-oriented orgs may prefer a “systematically irrational” approach:

Ideally only one alternative should be considered, so that no doubt arises about what should be done. Two alternatives could perhaps be taken up, but the second should then be so obviously inferior that it simply serves to enhance the attraction of the main alternative. Ideally, only the positive consequences of the alternative to be chosen should be described, so that it appears in the very best light, as a goal obviously worth realizing.

You may recognize this pattern from many corporate presentations: option 2 is described purely in negative terms; there’s no cost-benefit trade-off; or the rationale reads like we haven’t fully reflected stakeholder input. This kind of options analysis can sometimes feel less than objective or even clinical, as if it’s not faithfully representing the environment.And this is redoubled by the fact that LLM chatbots love to make this kind of options analysis, where pros and cons are rarely weighted by likelihood of occurrence or impact. But now we can appreciate that the sterility is by design! Option 2 exists solely to anchor the audience on option 1, which has effectively been preset as a function of the org’s ideological vector. The seemingly lacking text of the one-and-a-half-options analysis points to something that is in practical terms unwritable but that nonetheless haunts the PowerPoint slide, and that in fact reflects a coherent, principled organizational decision.

In this sense, the text of the options analysis is better viewed as ceremonial than as analytical. There’s a poetic tension here though: the participants in the ritual lose sight of the counterfactual,Wait, what would have happened if we’d seriously considered Option 2? whereas the outsider is armed only with a counterfactual that can’t reconstruct the ideology from the ceremony. Empowering the outsider is exactly how I recommended ungaming the counterweight, but it also illuminates the catch: you’ll never have leverage and standing at the same time, at least not by default. It’ll take either a keen observer or a strong cybernetician within to equilibrate these competing interests.

What does this mean for the wiki? Its paucity of content is just the image of an org that has already decided how knowledge ought to be shared, and how and when we ought to decide that effective sharing has been reached. Brunsson suggests that the action-oriented approach tends to “cultivate a spirit of enthusiasm”; but there’s little enthusiasm when the text you add surfaces doubts and works at cross-purposes with the org’s intent. On the other hand, the walls of text on WP:ANI reflect a more decisional, less action-oriented view of the question, one that springs from a more critical culture. And we see this most clearly in the Request for Comment (opens in new tab), where the options analysis that ensues must be started with a neutrally-worded question, lest it be considered procedurally invalid. The site is optimized for a precise kind of consensus decision-making, even if it might sometimes contradict what the community had settled on six weeks earlier. Neither approach is right or wrong in a vacuum; both are often precisely tuned to the org that produced them.

On things that aren’t worth doing

Brunsson’s philosophy of design helps us understand why your organization might act in ways that feel irrational from the ground floor. But while The Organization of Hypocrisy provides case studies that justify his read and largely defend against accusations of corporate cynicism, you may wonder how to apply this kind of academic diagnostic to your own org, especially if you don’t or can’t hope to change its ideology. A more pragmatic solution emerges from Gerald Weinberg’s timeless and very funny book More Secrets of Consulting. Weinberg inverts the traditional adage that anything worth doing is worth doing well, presenting instead what he calls “Cary’s Crap Caution”:Footnote 6Gerald Weinberg, More Secrets of Consulting, p 12-13.

Anything not worth doing is not worth doing right.

This is so simple as to seem obvious. And it’s largely true that if your time’s bounded, you’ll always need to cut in some places, so this rule just formalizes that ad-hoc process by setting a stopping rule somewhere past the inflection point on the logistic curve. Once you pass that point, the extra word isn’t worth writing. Now you might object that nobody who isn’t blessed with unusually clear foresight can know what might turn out to be important in the end. For one thing, if that were true, then it’d apply equally to the exhortation to do important things first, which nobody ever worries about.And if they do say it, they ought to read more Reinertsen. But the more practical counter is that we tend to work in iterations; what we don’t write in Act 1 is a trial balloon of sorts for whether we ought to course-correct later on.

Wikis are the canonical case because they’re designed with diffing in mind — so it’s not hard to see whether a short entry ends up causing problems. But a more daring example of the declining value of the marginal word is the kind of internal corporate tasking that, much like the wiki, has existence as its sole purpose. This doesn’t mean the compliance-directed ones that might spring to mind, but rather the ones that truly aren’t intended to be read, or at best in some interpassive (opens in new tab) sense, the documents read themselves. This makes them one of the few exceptions to the general prohibition I’d like to see on slop,Satire and programming the uninteresting bits are the others. I’m not a front-end web dev, but I’m also not interested in becoming one solely to publish my thoughts on this domain. a genre that threatens to upend a lot of the textual signalling we’ve discussed up until now. When it’s equally cheap to write one or one million lines of documentation, your org has to start selecting for something other than prose quality and volume to make decisions. You might largely restrict it à la Wikipedia,Footnote 7See Wikipedia:Writing articles with large language models (opens in new tab) (archived Jul 2026). or embrace it as a tool to write short options analyses even faster, but either way the old heuristics no longer apply.

So whether those decisions are made at the strategy retreat or the working level depends on your org, and to some extent how far away you sit from the locus of authority. On the wiki, I was well placed to issue-spot, but the action-oriented org is poised to filter exactly such critiques. Maybe the wiki was slightly detuned as a result, but changing that would have required something other than a few extra paragraphs of documentation.