Passive by Design (Sub: Spinoza, Enterprise Software, and the Illusion of Agency)

Every professional believes their workflow is an act of deliberate judgment. In reality, most of it was decided three years ago during an enterprise product review in Silicon Valley. Software vendors build for the widest possible market, which means their defaults, affordances, and data structures are engineered to make the generic way of working the path of least resistance.¹ The real sleight of hand is that it never feels like coercion. Whatever a menu makes easy feels natural, and whatever isn't on the toolbar simply ceases to exist in the professional imagination. Architects using Revit rarely imagining fluid, undulating spaces because the software insists on discrete storeys and datum levels; musicians in modern DAWs stop hearing natural tempo variations because the interface snaps every beat to an inflexible pixel grid.² By quietly deciding what can be represented and programmed, the tool ends up shaping not just our daily routine, but the final artifact, steadily nudging everyone's work toward the vendor’s commercial median.

A few centuries ago, Spinoza pointed out that human beings consider themselves free simply because they are conscious of what they want, while remaining blissfully ignorant of why they want it.³ Modern work run on off-the-shelf software is a textbook example of what he called passive: it is pushed along by an external cause. Autonomy here does not mean moving into the woods to draft plans with charcoal (an admirable fantasy, but difficult to invoice). Freedom means ensuring that the rules of your craft stem from your own internal necessity and distinct judgment, rather than the expansion targets of a software conglomerate.

When corporate software refuses to accommodate that distinction, the only viable defense is for a practice to build its own tools. Modern machine intelligence, primarily in the form of Large Language Models with programming abilities, makes this newly realistic. For the first time, a small firm does not need a massive enterprise IT department to construct disposable, single-purpose software. Traditional woodworkers have always understood this: when a standard table saw cannot handle an irregular angle, a carpenter does not redesign the furniture to be rectangular. He clamps down scrap wood and builds a jig, a rough, disposable device fitted directly to his specific cut. With language models, a practice can now knock together small digital jigs contoured precisely to its own idiosyncratic methods.

Naturally, there is a catch, because there is always a catch. Large language models are, by design, the ultimate consensus engines. They were trained on the internet, which means their default setting is to serve up the exact arithmetic average of everything humanity has ever typed online, including millions of corporate press releases, LinkedIn humblebrags, and furious forum debates about the right way to hang a toilet paper roll (it's over, never under!). If you leave a model to its own devices, it will not invent bespoke craft. It will politely smuggle the Silicon Valley vendor’s worldview right back through your prompt window while smiling pleasantly. The machine possesses infinite horsepower, but absolutely zero taste. It will happily pave over your firm's distinct instincts unless you stand over it and steer it with both hands.

Which brings us to the real dilemma: most people cannot actually steer it. Decades of cognitive research (conducted by scientists who apparently had to interview actual professionals for a living) have revealed an inconvenient truth: the better you are at your job, the less conscious knowledge you have of how you do it.⁴ A true master does not sit at their desk consulting a tidy, 14-step laminated checklist. Their brain simply looks at a disaster, fires off an instinctive neurological reflex honed by twenty-five years of screwing things up, and produces the right answer without bothering to notify their conscious mind.⁵ In fact, when researchers corner experts and demand an explanation for their brilliance, the experts routinely leave out up to seventy percent of what they actually did.⁶ Rather than admitting the truth ("I looked at the plan, felt an ancestral twitch in my left eye, and moved the doorway two feet to the left"), they panic and reach for textbook justifications about "maximizing spatial synergies," or they leave it at "that's just common sense" (which by the way, is not all that common).

So when a firm sits down to teach an AI its secret sauce, it unwittingly feeds the machine pure corporate platitudes, and then acts surprised when the machine hands back a piping-hot bowl of generic corporate oatmeal. The robot did not fail to listen; the firm simply could not bring itself to confess how it actually works.

The real challenge is therefore epistemological rather than technical, which is a fancy way of saying the problem is about what we know and how we know it. A digital jig is cheap. You can knock one together on a whim, use it to survive a frantic Tuesday deadline, and throw it in the virtual dumpster by Friday. But the craft knowledge required to conceive that jig in the first place is dangerously fragile.

When tools and workflows become disposable, the big hazard is no longer getting locked into an overpriced enterprise contract. The real hazard is institutional amnesia. Every time a seasoned practitioner packs up their favorite mug and walks out the door, decades of unwritten, instinctive judgment go down the elevator with them. The surviving team isn't just left staring at an abandoned template or a checklist nobody understands. The firm has lost the living circulatory system of its craft: its private definitions of quality, its prejudices, and the courage to reject the vendor's average. Software and systems can be replaced in an afternoon; the taste that steered them took a generation to build.

Practices that take the trouble to pry those principles out of their experts' heads can treat their software as totally expendable matter: building a jig, using it, tossing it out, and letting their underlying standards endure. Practices that fail to do so will simply trade one Silicon Valley master for another, drifting straight into the generic average while congratulating themselves on their cutting-edge independence.

So before you sign another year-long enterprise software contract, consider this: corner your grumpiest, most seasoned veteran in a conference room, lock the door, ply them with high-end pastries, and refuse to let them leave until they explain (in plain English, with zero buzzwords allowed) the exact, visceral reasons they hated the last five proposals that crossed their desk. Write down every twitch, every unwritten rule of thumb, and every hard-won prejudice. That messy, unlaminated instinct is the only thing keeping your firm distinct. Build your disposable AI jigs, your templates, checklists, written documentation, and whatever strange things you require around that, or get comfortable churning out lukewarm corporate oatmeal for the rest of your natural life.

Footnotes

1. Donald A. Norman, The Design of Everyday Things, rev. ed. (New York: Basic Books, 2013). Norman formalizes how interface affordances and constraints dictate human behavior by making certain actions intuitive while rendering alternatives invisible. ↩

2. For the architectural impact of parametric databases, see Mario Carpo, The Alphabet and the Algorithm (Cambridge: MIT Press, 2011). On how digital audio workstations and grid-quantization alter compositional habits, see Mark Katz, Capturing Sound: How Technology Has Changed Music, rev. ed. (Berkeley: University of California Press, 2010). ↩

3. Baruch Spinoza, Ethics, ed. and trans. Edwin Curley (Princeton: Princeton University Press, 1985), Part III, Proposition 2, Scholium; see also Spinoza’s Letter 58 to G. H. Schuller (October 1674), which uses the famous analogy of a stone propelled through the air believing it flies by its own free will. ↩

4. Michael Polanyi, The Tacit Dimension (Chicago: University of Chicago Press, 1966), 4. Polanyi summarizes the dilemma in his famous aphorism: "We can know more than we can tell." ↩

5. Hubert L. Dreyfus and Stuart E. Dreyfus, Mind over Machine: The Power of Human Intuition and Expertise in the Era of the Computer (New York: Free Press, 1986). The Dreyfus model demonstrates that true experts abandon conscious rule-following in favor of intuitive, holistic situational recognition. ↩

6. Richard E. Clark, Kenneth Yates, Sean Early, and Kimberly Maxwell, "Cognitive Task Analysis," in Handbook of Research on Educational Communications and Technology, 3rd ed., ed. J. Michael Spector et al. (New York: Lawrence Erlbaum Associates, 2008), 575–593. Clark documents that domain experts unconsciously omit roughly 70 percent of procedural steps and decision criteria during self-reporting. See also Richard E. Nisbett and Timothy D. Wilson, "Telling More Than We Can Know: Verbal Reports on Mental Processes," Psychological Review 84, no. 3 (1977): 231–259. ↩