Why Realistic Quality Standards Matter More Than Ambitious Ones

Think about the last time a product failed you in a way that felt entirely preventable. A medical device that glitched. An app that crashed during something urgent. A piece of software that quietly mishandled your data. Chances are, somewhere in that product's development, a quality standard existed on paper that nobody could actually meet in practice. That gap between what's written down and what's achievable is where real harm tends to live.

This is the core insight behind a widely discussed IEEE Spectrum piece on product management, part of a broader list of ten practical lessons for building better technology. The advice sounds almost too simple: insist on realistic quality management. But simple doesn't mean easy, and it certainly doesn't mean common. Too many organizations, whether they're building consumer electronics, medical devices, or increasingly, AI systems, set quality benchmarks that look impressive in a boardroom presentation but collapse under the weight of real-world engineering constraints, budgets, and timelines.

Here's the plain-language version of the problem. Imagine a fitness coach who tells you to run a marathon next week when you've never jogged a mile. The goal isn't wrong in principle, marathons are achievable, but the timeline and preparation make failure almost certain. Quality standards in product development often work the same way. A team commits to a defect rate, a safety threshold, or a compliance target that sounds rigorous, but without the resources, testing infrastructure, or staffing to actually hit it. The standard becomes theater. Everyone nods along in meetings, and then the product ships anyway, gaps and all, because the deadline doesn't move even when the standard was never realistic to begin with.

The Governance Gap Nobody Talks About

The consequences of this gap ripple outward in ways that matter well beyond the engineering team. When quality management exists mostly as a checkbox exercise, the people who bear the cost are the end users, the patients, the drivers, the customers who trusted that "meets quality standards" meant something. This is especially urgent as AI systems get embedded into higher-stakes environments: hiring tools, medical diagnostics, autonomous vehicles, financial risk models. An AI standard that's aspirational rather than enforceable doesn't just fail on a spec sheet. It fails somebody's job application, somebody's diagnosis, somebody's loan.

Governance frameworks are supposed to close this gap. That's their entire purpose: translating good intentions into enforceable, testable requirements. But governance only works if the people setting the standards are honest about what's achievable given real staffing, real budgets, and real timelines. A framework that demands zero-defect AI outputs, for instance, sounds admirable until you remember that even well-tested AI systems can behave unpredictably in edge cases nobody anticipated. Setting an impossible bar doesn't raise the actual quality of the product. It just teaches teams to quietly game the metric, or worse, to stop taking the standard seriously altogether.

There's a useful parallel here to public health policy, where I've spent much of my career. Health regulators learned decades ago that setting unattainable targets, like demanding zero workplace injuries with no plan for how to get there, tends to backfire. Workers and companies alike start treating the goal as symbolic rather than operational. The standards that actually reduce harm are the ones built with input from people doing the work, tested against real conditions, and revisited when circumstances change. Product quality management needs that same humility. A standard nobody can meet is not a high bar. It's an invisible one, because it exerts no real pressure on anyone's actual behavior.

This doesn't mean lowering ambitions. It means separating aspiration from obligation. An organization can absolutely aim high, publicly commit to continuous improvement, and push its engineers toward excellence. But the enforceable floor, the thing that actually gets audited, tested, and tied to accountability, needs to be something a competent team can hit given the resources they actually have. When that floor is set honestly, it becomes a tool people trust. When it's set aspirationally, it becomes a tool people learn to route around.

The comments beneath the original IEEE Spectrum piece hint at a related cultural issue worth naming directly. One reader pushed back sharply against the idea of promoting people who consider hands-on quality work "beneath" them, arguing instead that such attitudes should get someone removed from a team entirely. It's a blunt point, but it lands on something real. Quality management fails not just when standards are unrealistic, but when the people responsible for meeting them view the work itself as low-status. If your best engineers see rigorous testing and defect tracking as grunt work to escape via promotion, you've built an organizational incentive structure that actively erodes the very quality controls you claim to value.

Why It Matters

None of this is abstract for the AI governance conversations happening right now. Regulators, standards bodies, and companies are actively debating what "quality" and "safety" should mean for AI systems being deployed at scale. If those definitions get written the way many product quality standards have historically been written, ambitious on paper, disconnected from engineering reality, we'll end up with the illusion of oversight rather than the substance of it. Realistic quality management isn't a modest ambition. It's the difference between a standard that protects people and one that simply looks good in a press release.

Getting this right requires patience that policy conversations don't always reward. Realistic standards take longer to negotiate because they demand honest conversations about resource constraints, technical limitations, and genuine tradeoffs. That's slower and less satisfying than announcing a bold target. But bold targets that nobody can meet don't protect anyone. They just delay the reckoning until a product fails in the real world, at which point the people who suffer aren't the ones who wrote the standard. They're the ones who trusted it.