Clay Shirky (“The Key to Successful Tech Management,” March/April 2014) makes so many good points that it seems petty to disagree with just a few. But the stakes are life and death when it comes to getting his management argument right, especially as it regards health care and the Affordable Care Act.

Health services, including those delivered online through websites such as, are mandated to be very reliable in their safe and continuous provision, even when demand for those services is high. “Embracing failure” may well work in constructing “novel infrastructure,” but I know of no critical infrastructure for which the control room embraces failure in real time. True, the operators -- be they in hospital emergency rooms or the control centers of major electricity grids -- face tradeoffs, take risks, innovate, and manage the unexpected, but they do so to reduce the risk of failure. The reliability professionals I study are always running scared from failure -- which, I suspect, is precisely what the management consultant Jeffrey Zients and his staff were doing when they were tasked with fixing

Shirky is too ready to see all virtue in learning by doing and no virtue in planning ahead. In reality, reliable managers must do both: they have to be resilient in bouncing back from the inevitable surprise and take steps to avoid system breakdowns.

The urgent issue is not so much how to reduce the rate of failure of new technology but rather how such a reduction will actually improve the reliability of health care. For the uninsured, any improvement is an improvement. But the last thing they, like the rest of us, would want an airplane pilot to do is to embrace failure. And the same goes for the health-care navigators connecting people to insurance plans that could save their lives.

Associate, Center for Catastrophic Risk Management, University of California, Berkeley

Shirky Replies:

As Emery Roe himself notes, there are tradeoffs even in situations far more critical than the provision of health insurance. No amount of advance planning can guarantee success in all conditions. Yet in considering the failures of, he makes the very mistake I document in my piece: refusing to believe in the possibility of a phased rollout -- even after one has occurred. The single most salient fact in the entire debacle is that two years of mandates for high reliability and continuous service did not produce high reliability or continuous service.

Those who rescued the site after its disastrous launch assumed that different sorts of failures were of different levels of importance and triaged them. This meant treating some forms of failure as acceptable and all forms of failure as diagnostic. It also meant regarding nontechnical edicts as less relevant than facts on the ground.

And contrary to Roe’s lauding of hospital culture as a counterexample, we now know that decrees to avoid medical failure led to decades of systematic underreporting of errors, including fatal ones. Managerial correctives to opaque hospital culture have included such things as systematic mortality reviews, which require exactly the sort of acceptance of failure that allows hospital culture to adapt to the lessons of failure.

Managers love to believe that they can mandate failure away. They cannot. They can, however, create circumstances in which failures are small, early, and recoverable, rather than big, late, and catastrophic. The team that worked on for two years did not create these circumstances. The team that rescued it in two months did.