People Are Not Interchangeable Units
One of the most persistent and damaging assumptions in engineering management is that people are fungible. Roadmaps are drafted in terms of "heads" — one person for one quarter, two for three quarters — as though headcount were the only variable that matters. The polite fiction of the planning process is that anyone can do anything, and that the identity of the person assigned to a project is irrelevant to its outcome.
In practice, everyone downstream of that planning knows it isn't true. When teams are asked to depend on another team's deliverable, the first question is almost always "who is actually doing the work?" Concrete, granular familiarity with individuals — their effectiveness, history, and capacity — overrides the roadmap's abstract assumptions. Dependencies are planned around whether a given person or team can realistically pull off a project, not whether the roadmap says it's staffed.
The problem is not just theoretical forecasting error. Individuals on a team accumulate institutional knowledge and shape its working culture. That context is hard to build ex ante and easy to destroy. As the author notes, "an effective team is difficult to create... but destroying a team is very easy." The human cost of treating project assignments as if people were interchangeable playing pieces has been demonstrated repeatedly in practice.
When the Bureaucracy Prizes Headcount Over Talent
This assumption propagates beyond roadmapping and into how companies are actually financed and run. Budgeting processes at many non-engineering organizations and finance teams force engineering to convert dollars into "heads," with little flexibility to treat people as genuinely different in value. A budget item for "three staff-level heads" cannot be traded for "two more effective and better-paid staff-level heads" without a fight from finance.
The result is that hiring managers often find themselves unable to bring on the one candidate they know is worth ten of a generic head. HR or "comp teams," whose mental model of an employee consists of role, level, and location, object that a candidate is too expensive for their level. The bureaucracy is designed so that no one with real understanding has authority to override a standard rate card, favoring procedural fairness and variance reduction at the expense of overall effectiveness.
This mindset shows up in retention crises too. The article recounts the case of an engineer who added meaningful revenue for his company — checking multiple claims, the author says, he found this person's wins credible, including a project completed while on paternity leave that increased the company's revenue by 0.7% — and whose request to relocate to Portugal was met with a scheduled fourfold pay cut (or a double cut if he chose Spain). The company, unwilling to override its location-based pay formula, lost a peerless performer.
Attrition Knobs Cut Both Ways
The same logic that prioritizes cost and title over individual value drove two companies' HR departments to judge their attrition rates as "too low" and deliberately provoke more losses. Of course, attrition induced by policy does not sample uniformly across the talent distribution. In one case the price of moving the overall rate from the "unhealthy" ~5% to a manager-approved threshold was mass loss among exactly the engineers the company relied on most.
The author reports the sharpest effect obliquely: in the case he saw closest, the most effective engineers quit at twice the company's elevated average rate — at least, that's understating it, since those losses were disproportionately concentrated in long-tenured senior staff, where normal expected attrition is a tiny fraction of the mean. That sort of severe right-tail damage, cutting off the long tail of extreme outlier value in exchange for the comfort of a legible knob diagram, is effectively impossible to repair.
Applying the Same Lens to the Social and Aid Sectors
The same mental error, discounting individual human agency in favor of treatable-average assumptions, afflicts effective altruism and development work as well. The author's case study is a three-person team working on "the ground" in African countries helping reach an ignominious benchmark — near-eradication of guinea worm in a region. At a $12/(person-day) budget, the intervention's returns are off the charts in retrospect. Yet an experimental, cleanly quantifiable "proof" for such a move is structurally impossible: the value lies in team chemistry, personal judgment, and tactical context, none of which is distributed like a standard intervention in a randomized trial.
The guinea worm victory was the self-contained, measurable side of the team's value. The harder-to-quantify, larger payoff was in advising mid-level government officials on how to place water infrastructure more wisely, and hence supporting better decisions across many types of public investment. On an aggregate expected-value basis, the team was likely one of the highest-ROI units in development anywhere — but the "many people want the world to be simple" crowd, and even the leadership within their own funding organization, tried to pull resources toward interventions whose payout could be computed with greater ease, either because computation was itself simpler, or because the standard-bearers of the approach (like an EA-aligned friend at GiveWell) had to earn initial credibility through inexpensive, legible wins.
That is arguably a rational way for a fledgling organization to build credibility — the author does not dispute it, and he reports that a friend inside GiveWell at the time of its early years maintains that the focus was neither overvaluing easy metrics nor undervaluing messy org work. But when donors or managers apply the heuristic as a rigid all-or-nothing rule — "charity on GiveWell list = good; not on GiveWell list = bad" — they outsource their judgment to a simple metric and pay an enormous cost in efficiency.
The High Cost of Scalable Thinking
Watching "leaders persist in thinking they can re-direct people" as though swapping tasks around were like re-allocating workers in a city simulator, the author highlights common disasters. One high-visibility, public-facing project was canceled weeks from launch after a year's investment, then un-canceled and revived as a diminished initiative staffed by new graduates under a redrawn org chart. A manager who tried to report that the team just wanted enough runway to finish was overruled. The entire original team quit; the surviving project's decayed remains were the sustained output for years afterward. Another reorg attempted to force six volunteers onto a low-credibility project. Result: two senior devs quit, an engineering manager retired, a PM was fired, and the project team had to be rebuilt from nothing.
The central theme is that the individual perspective is practically the only thing that forecasts code-team success or organizational behavior — project-assignment warnings from individual contributors in senior roles are reliably prognostic. Yet leaders instinctively reach for the abstraction layer they can scale, standardize, and read on a single slide. When returns are dispersed evenly, the optimization from forced legibility may not cost much. But in a heavy-tailed, winner-take-most market, promoting "lowest-common-denominator effectiveness" is not merely sacrificing optional upside: per the author, turning those knobs can "severely tamm down the right side of the curve" and cost the majority of possible gains. The goal is never legibility alone — it is reading the nuances of who is on your team and taking their individual edge seriously. If org-building becomes reducible to a software patch that has no better vision of people than a head count of uniform units, that efficient-looking sameness eventually becomes the drift that causes the committee to form.
