TBM 437: AI and the Recontextualization Tax

John Cutler · email · 2026-09-03 Reader 🖥️

Agent Skills Cognitive Debt Cognitive Load Organizational Abstraction Levels Organizational Legibility Proxy Validity (Construct Validity of Measurement Proxies)

Organizations pay a real, often-hidden cost -- the 'recontextualization tax' -- every time people flatten a working reality into a simplified version to suit whoever has the most power, the least appetite for detail, or the least time, and AI is newly good at absorbing that translation labor because it can render one underlying, accurate context into many audience-tailored formats without forcing teams to first standardize how they actually work.

practical · expository — A short, single-thesis newsletter post: names a recurring organizational cost (the 'recontextualization tax'), diagnoses its cause (flattening reality for the least-detail-tolerant, most powerful, or most time-starved audience), and proposes AI-as-translator as a practical remedy. No formal argument structure or narrative arc -- it is a compressed op-ed built from one extended example (the 10-stakeholders/10-roadmap-versions case) plus a closing personal application.

Structure

Part	Role
Naming the pattern: companies flatten reality for power/low-detail-appetite/time-poor stakeholders	sets up the problem
The CEO's 'too in the weeds' moment forces a binary: invest in flattening, or wing it	concrete triggering scenario
Naming the tax explicitly and noting it can be so large that orgs choose to wing it rather than pay it	thesis statement / key term introduction
Reframing the demand for simplicity as a legitimate cognitive-load survival mechanism, not pettiness	charitable reframe of the powerful stakeholder's behavior
Pivot: 'Which brings me to AI' -- AI is good at the grunt work of recontextualization and exception-hunting, given accurate/current raw context	solution claim
10-stakeholders example: pre-AI you either persuade everyone to use one format or hand-build N versions; this is why roadmapping tools don't travel	narrative/illustration
AI-as-skills-per-audience: let a team work however it wants and write a translation skill per audience segment	proposed mechanism
Personal application at 'the new job': skills that translate internal working reality to each client's preferred language, without forcing clients or the team to converge on one format	closing narrative / self-application
Closing maxim: AI's value here is un-blocking things you already knew you should do but lacked hours for	generalizing coda

Key terms

Recontextualization tax
The time/effort cost paid -- by a few people or by many, across an org -- when a working reality has to be translated, simplified, and flattened into a different format to suit an audience with more power, less appetite for detail, or less time than the people who produced the original information.
Flattening Organizational Legibility
The act of stripping detail and nuance from a working reality to produce a simplified representation (a report, roadmap, or overview) legible to a less-detail-tolerant audience.
Managing by exception
The coping strategy of powerful, time-poor stakeholders: rather than viewing full complexity, they want a simplified view that only surfaces deviations/exceptions worth their attention.
Grunt work of recontextualization
The repetitive, low-judgment translation labor of restating the same underlying information in different formats for different audiences -- the specific labor Cutler says AI is well-suited to absorb.
Skill (per-audience translation skill) Agent Skills
A reusable AI capability/prompt tailored to one audience segment's preferred vocabulary and mental model (e.g. swimlane roadmaps vs. outcome-metric framing), used to translate a team's actual working reality into that audience's preferred format on demand.

Propositions

explicitOrganizations routinely flatten information to fit the person with the most power, the least appetite for detail, and/or the least time.conf: high
extendsTBM 414: Legibility and Legitimacy · p1 — This essay's 'flattening for the powerful/detail-averse/time-poor' is a concrete instance of tbm-414's claim that legibility requires selective simplification enabling control over those being simplified -- here Cutler proposes AI as the agent performing that selective simplification, without revisiting the power dimension he himself flagged in tbm-414.
explicitWhen a powerful stakeholder rejects a report as 'too in the weeds,' the organization is forced into one of two responses: invest heavily in flattening reality to the 'right' level, or skip that investment and improvise ('wing it') at the last minute.conf: high
supportsTBM 437: Tokens, Hours, Points, and Other Curious Proxies · p9 — Both pieces (published in the same TBM 437 issue) share Cutler's broader concern with organizations substituting an easy/legible proxy (a flattened report, a token metric) for the harder work of maintaining an explicit, faithful model of underlying reality/value.
explicitcontestedThe cost of translating one version of organizational reality into a simplified, flattened version -- the 'recontextualization tax' -- can be large enough that organizations choose to forgo it and improvise instead, rather than pay it.conf: high
contradictsTBM 434: How Maps Can Hide Problems · p8 — tbm-434#p8 warns that AI smoothing over organizational contradictions 'can exacerbate the underlying tensions while producing a false appearance of alignment.' This essay praises the structurally similar act of AI smoothing/flattening reality into simplified, audience-pleasing formats as a good thing, without acknowledging the risk his own earlier essay raised about the same mechanism -- a live tension between the two pieces rather than a clean resolution.
explicitThe demand for simplicity from high-level stakeholders is, in most cases, a legitimate response to a real cognitive load constraint rather than pettiness or laziness.conf: high
explicitAI is well-suited to performing the grunt work of recontextualizing information and hunting for exceptions, provided it has access to raw, reasonably up-to-date context.conf: medium
extendsTBM 434: How Maps Can Hide Problems · p7 — tbm-434 argues AI cannot reconcile incoherent organizational maps and can only surface/compare them; this essay extends that by proposing a narrower, more tractable task AI CAN do well -- translating one coherent underlying reality into multiple audience-specific formats -- implicitly distinguishing 'translation of coherent reality' from 'reconciliation of incoherent structures.'
explicitRoadmapping tools rarely spread successfully across organizations because different stakeholders/audiences need different things from a roadmap, making ad hoc, per-audience presentation-building (e.g. in PowerPoint or Miro) easier than adopting a shared tool.conf: medium
explicitAI makes it newly practical to let a team work however is genuinely effective for them while still producing an artfully tailored, audience-specific translation of that work for each stakeholder segment, via a dedicated translation 'skill' per audience.conf: medium
implicitIt is preferable for an organization to build audience-specific translation skills than to force every client or stakeholder to adopt one shared working format, and preferable to that than over-customizing an internal tool to accommodate every external workflow.conf: medium
explicitA significant share of AI's practical value comes not from novel high-end use cases but from finally doing things people already knew were worth doing but never had enough hours to execute.conf: high

Arguments

AI is (for now) a genuinely good tool for absorbing the recontextualization tax that organizations otherwise pay in wasted time or forgo by winging it. strength: medium

Explicit premises:

Organizations pay a real cost (the recontextualization tax) translating one working reality into simplified, audience-specific versions.
That cost is driven by legitimate constraints (power, cognitive load, time) on the receiving audience, not by pettiness.
AI excels at the grunt work of recontextualization and exception-hunting, given accurate and current raw context.
AI enables building a distinct translation skill per audience segment without requiring the underlying team to change how it actually works.

Implicit premises:

The 'raw context' condition (accurate, up-to-date) is realistically achievable in most organizations wanting to use this approach.
A faithful translation into a different format does not itself introduce new distortions or omissions beyond the ones already accepted as the cost of flattening.
Stakeholders will trust and accept an AI-mediated translation of reality as readily as a human-produced one.

Inferences:

Because the tax is proportional to the number of distinct audience formats needed, and AI marginal-cost-per-additional-format is low, AI structurally changes the economics of the tradeoff between standardizing everyone's format and hand-building N versions.

Narratives

The CEO who says 'This is too in the weeds. There's no way we are doing this much!'
→ Illustrates the triggering event that forces the choice between investing in flattening or winging it; personifies the 'least appetite for detail / least time' stakeholder.
10 stakeholders, each wanting information 'a certain way,' and the pre-AI choice between persuading them to converge or hand-building 10 versions of every presentation.
→ Demonstrates the scale problem the recontextualization tax creates and sets up why AI-generated per-audience translation is valuable; also explains roadmap-tool failure as a side effect.
Cutler's own plan at 'the new job': building skills that translate the team's actual way of working (GitHub issues, Google Docs) into whatever language each client prefers, without hyper-customizing Linear or forcing client conformity.
→ Self-application narrative that grounds the abstract thesis in a concrete, current first-person project, modeling the recommended practice.

Questions addressed

✅ Why do roadmapping tools rarely spread successfully across an entire organization?
Because different stakeholders/audiences need fundamentally different things from a roadmap, making ad hoc per-audience artifacts (PPTX, Miro) easier than adopting one shared tool.
✅ Is the powerful stakeholder's demand for a simplified view a legitimate need or an unreasonable one?
Legitimate -- it reflects a real cognitive load constraint and a 'manage by exception' survival strategy, not pettiness.
⭕ Can AI-mediated recontextualization introduce its own distortions, omissions, or loss of nuance -- i.e., does the 'tax' actually get paid in full, or just hidden?
⭕ What happens when the 'raw context' AI needs is not actually available, current, or complete -- does the recontextualization-tax solution degrade gracefully or fail silently?

Gaps

incompleteCutler names one precondition for AI paying the tax well -- 'it has the raw context and that context is reasonably up to date' -- but never explores what happens when that precondition fails: stale, partial, or politically curated context would presumably make AI-generated translations confidently wrong rather than usefully simplified, and the essay gives no signal for how to detect that failure mode.
Read next: Compare with Cutler's own tbm-434 (extract-tbm-434-how-maps-can-hide-problems), where he argues AI cannot reconcile genuinely incoherent organizational structures and, if it smooths over contradictions rather than exposing them, can 'exacerbate the underlying tensions while producing a false appearance of alignment' -- the same smoothing mechanism praised here as a feature is treated there as a serious risk. The two pieces would benefit from being read together to see whether Cutler distinguishes 'translating a coherent reality for different audiences' from 'smoothing over an incoherent one,' since from outside his own framing the difference may not always be legible in practice.
incompleteThe essay treats the recontextualization tax purely as a cost to be minimized/automated away, without asking whether the act of flattening/translating reality for a powerful audience ever serves an epistemically or organizationally valuable function beyond accommodation -- e.g. forcing someone to compress a complex reality into a one-page summary can itself surface what actually matters, a benefit lost if AI does the compression silently on the recipient's behalf.
Read next: Cross-reference tbm-414's treatment of legibility and power (extract-tbm-414-legibility-and-legitimacy#p1: 'making systems legible requires selective simplification that enables control over those being simplified') -- if flattening is partly about control, not just cognitive load, then having AI perform the flattening doesn't remove the power dynamic, it just launders who is doing the (potentially distorting) selection.
uninformedNo engagement with what happens to accountability when a stakeholder acts on an AI-translated version of reality that later turns out to have mis-flattened something important -- who is responsible for catching that, and does the translation layer diffuse responsibility in a way that hand-built flattening (where a named person chose what to cut) did not.
Read next: Read alongside Cutler's own cognitive-surrender / cognitive-debt material elsewhere in the corpus on AI-mediated understanding gaps, to see whether he has a general position on accountability for AI-intermediated organizational communication.