AI is accelerating a management mistake we've made before

When organizations optimize top-down for measurable outcomes, they systematically erode the ground-level judgment, craft, and cross-functional thinking that complex systems actually run on.

Organizations have more specialists than they did a decade ago. They also have more coordination overhead, more handoffs, and more process to manage the gaps between them. The promise was efficiency. What often arrives instead is brittleness; a quieter loss of the organizational adaptability that no dashboard measures.

The product management role was similarly generalist in the near past. When I started my career, I was a Swiss Army Knife. I generated ideas, validated them, submitted them for approval, wrote their specifications, managed their build, tested their quality pre-release, and marketed their features and provided customer support post-release. Today, that same scope may be distributed across a Product Strategist, a Go-to-Market Specialist, a User Researcher, a Business Analyst, a Scrum Master, and a Product Marketer — each reporting to a different head of department. The specialization was meant to create more value. What it created instead is worth examining.

The same pattern played out in healthcare with the move to digitize medical records. Dr. Atul Gawande's New Yorker account of first-generation Electronic Medical Record (EMR) describes an overdue digital transformation. But it's really about what happens when top-down design decisions optimize for the people giving orders rather than the knowledge workers who deliver service value.

How we got here

The organizational adaptation to new technologies like AI agents, low-code platforms, and specialized software roles didn't emerge from nowhere. It follows a logic that dates to the Industrial Revolution when manufacturing work was first broken into components of standardized parts — in a process we now call Taylorization. Designers were separated from builders, and workers were optimized for narrow, repeatable tasks. The efficiency gains were real, but so were the tradeoffs.

Dr. Gawande describes what those tradeoffs look like when the logic is applied to complex knowledge work:

Adaptation requires two things: mutation and selection. Mutation produces variety and deviation; selection kills off the least functional mutations. Our old, craft-based, pre-computer system of professional practice—in medicine and in other fields—was all mutation and no selection. Computerization, by contrast, is all selection and no mutation. Leaders install a monolith, and the smallest changes require a committee decision, plus weeks of testing and debugging to make sure that fixing the daylight-saving-time problem, say, doesn’t wreck some other, distant part of the system.

The cost of executive-first design

When resources are allocated top-down toward executive goals of control, oversight and reporting, the administrative burden flows downward onto the worker. EMR software gave hospital leadership powerful tools to monitor population health and enforce compliance, but its lack of consideration for doctors didn't mitigate burnout at all.

I experienced the same dynamic when leadership mandated a program management tool designed to push standardized progress updates to VP dashboards. The goal was executive visibility. The result was product managers making a dozen additional clicks to update a single administrative field that yielded no better outcomes for the organization, and quietly eroded the time available for the work that actually created value.

This is what I call trickle-down software: technology that serves the top of the hierarchy at the expense of the people tasked with creating value. The assumption is that better executive visibility produces better organizational decisions but, as Gawande observed, the more consequential cost is what gets lost in the process:

Artisanship has been throttled, and so has our professional capacity to identify and solve problems through ground-level experimentation... Technology for complex enterprises is about helping groups do what the members cannot easily do by themselves—work in coördination.

The risk that can't be measured

In reality, every complex system contains what Taylor Pearson calls 'illegible margin', which is the judgment, improvisation, and ground-level knowledge that doesn't appear in KPIs but quietly determines whether the system thrives or stagnates over time. This explains why highly planned cities like Brasilia underperform the organic complexity of unplanned ones like New York's West Village. In centrally-planned communities, urban life is limited to only account for what the planner anticipated.

First-generation EMR demonstrated this at scale. Its brittleness didn't just frustrate doctors; it forced the medical system into a patch that illustrated the root problem. Hospitals began hiring human 'medical scribes' to enter and retrieve data that the software was supposed to handle. As Gawande noted: "this fix is, admittedly, a little ridiculous. We replaced paper with computers because paper was inefficient. Now computers have become inefficient, so we’re hiring more humans."

The downstream consequences compounded. A single thirty-minute patient visit required an hour of offshore scribe processing, review by a second physician, and sign-off from an insurance coding expert. Scribes carried high error rates and turned over within months. Physician dissatisfaction, stress, and depression climbed as both the highly paid specialist and the minimum wage worker were bearing costs the executive dashboard never captured.

Rather than questioning the underlying design philosophy, the medical system has kept iterating by replacing human scribes with AI scribes. However, as one reviewer concluded, doctor recollection of visits has suffered and the conversation with the patient changes in unforeseen ways when it's being recorded. And, when an AI records everything indiscriminately, the clinical summary risks burying the most medically important information under a volume of equally-weighted detail — detail that the clinician must now demonstrably address because it was recorded.

More fundamentally, the doctor's desire to help and treat patients — which has been obfuscated in ways that drives burnout — isn't improved by automations.

What specialization costs

Gawande traced physician burnout to the pointlessness of the administrative work that displaced meaningful clinical judgment. The same dynamic appears in knowledge work. When a role is narrowed to a repeatable set of tasks, the practitioner loses scope, agency, and the motivation that comes from owning something end to end.

I've watched this play out among product management colleagues. Specialization has promised clarity of ownership, but what many teams quietly report is more coordination cost, more predictable output, and fewer people asking the uncomfortable questions that sit between job descriptions.

AI in software development promises to reduce specialization by collapsing design, product, and engineering roles, but this promise is undercut by most job listings advertising "AI" as the job function. To restrict knowledge workers to the use of a specific tool — rather than to service a customer, a problem, or a market — is to embrace specialization of a different stripe and ignore the illegible margin.

The subtler cost is what happens to the people in those narrowed roles over time. Specialists optimized for task execution gradually lose — or never develop — the capacity for the kind of cross-functional, systems-level thinking that organizations actually need when things go wrong or markets shift. What looks like a more efficient organization is often a more fragile one.

A different path

This isn't inevitable. Postwar Japan and West Germany demonstrated that the alternative to Taylorized specialization isn't inefficiency; rather, deliberate investment in autonomous workers creates durable value. Their approach to continuous improvement, which trained and enabled front-line workers to identify and solve production problems rather than escalate them upward, outperformed American top-down management on both quality and costs by the late twentieth century.

The "both/and" path exists. Top-down coordination and bottom-up problem-solving aren't mutually exclusive. Agile software development was designed partly on this principle. Its mixed application in practice says less about the principle than about how rarely organizations implement it with genuine worker autonomy.

For leaders navigating the current wave of AI adoption, the question isn't whether to use the tools. It's whether the implementation empowers the people doing the work or further removes their judgment from the process. Every new AI-powered workflow, every automated dashboard, every specialized agent inserted into a knowledge work process deserves the same scrutiny: who bears the cost of this efficiency, and what judgment are we quietly removing?

The most enduring value of breakthrough technology, including AI, is not in its ability to maximize output in the short term, but in its capacity to elevate the human worker to maximize quality, innovation, and sustained success.

Get email updates on the latest from Do this well

Leave comments, questions or requests