The ground stop was the plan. And that's the problem.
On the afternoon of November 13, the Federal Aviation Administration (FAA) halted departures at every airport in the United States. Communication problems had hit the Air Route Traffic Control Centers, the facilities that guide aircraft between airports at cruising altitude. About 3,100 aircraft were already in the air and bound for U.S. airports.
As I write this, the FAA hasn't named a root cause. As a certified flight instructor, I'm as curious as anyone. But for most organizations, the root cause isn't the interesting question. The interesting question is what happened next, and what it cost.
The fallback worked. It still had a cost.
Let's be clear: the ground stop worked. Every flight was accounted for, and no new aircraft entered a system that couldn't safely handle them. Safety came first, as it should.
But a ground stop is all or nothing. Departures drop to zero, delays ripple across the country, and the cost lands on airlines, passengers, and every business waiting on them. The gap isn't between having a plan and having none. It's the missing middle: a degraded mode that keeps part of the system running while the failure is fixed.
That middle usually goes missing for familiar reasons. Someone once judged the scenario unlikely. Full redundancy looked too expensive. Or the backup existed but was never exercised under real load. This isn't a criticism of the FAA. Most organizations make the same trade-offs, often without realizing it.
Emergency managers have a model for this
Emergency managers have worked on this problem for decades. The Federal Emergency Management Agency's (FEMA) disaster cycle runs through mitigation, preparedness, response, and recovery, and it never ends. Each pass draws on what has happened before and on a forward-looking view of what's likely next. Each pass also leaves you better prepared.
Their approach is all-hazards. Disruption can come from nature, such as weather. It can come from people acting deliberately to cause harm. It can also come from an unplanned or untested change made by someone who couldn't see what it would break.
Why IT continuity plans fall short
Most organizations already have disaster recovery, major incident management, and business continuity plans. The trouble is that the first two are usually treated as technical exercises: restore the server, patch the software, and find the root cause.
Root-cause analysis matters, but it doesn't land the plane. A failed shopping cart checkout is costly. An aircraft at 35,000 feet with finite fuel, possibly flying into weather without radio contact, is something else. Pilots train for that, but it still cuts capacity across the whole system and raises risk and cost for everyone.
And continuity isn't only IT's job. Technology may own the systems, but customer service, operations, finance, and communications all have a part in keeping the business running. Making it IT's problem assigns blame. It doesn't create a plan.
We shrink the possible to the probable
Why don't plans cover more? Because the space of what could go wrong is vast, and people don't handle vast spaces well. We sensibly shrink the possible down to the probable and focus there. Fatigue, pressure, and the limits of what anyone can hold in mind make that necessary.
Adding more people doesn't solve it. The area where everyone agrees shrinks, so the scenarios a group commits to can narrow instead of grow. In this sense, people don't scale well.
An AI partner can hold more of the whole
This is where AI partners change the math. Most people think of AI as a tool: ask a question, get an answer, move on. That's useful, but it isn't a partnership. An AI partner is a working relationship. You give it context about your business, work through a problem with it over time, question its reasoning, and stay accountable for the result, much as you would with a capable colleague.
A partner like that doesn't tire, and it can keep far more context in view than any one person. That lets a team consider scenarios it would otherwise set aside, including the unlikely ones.
AI partners don't replace people, and they make mistakes. The division of labor is simple: let the AI partner widen the field, and let people judge which scenarios matter, what they would cost, and what to do about them.
Nor does this mean buying a second copy of everything. Work through the problem in chunks, one service at a time, and build the broader picture as you go.
From seeing to acting
This is also where SuperOps comes in. We build endpoint management technology, and the problem in this article is one we work on every day.
IT has never had more visibility. There's more monitoring, more alerts, more dashboards, and more data than ever. Yet teams are still expected to connect the dots, understand the consequences, and make the right call before something goes wrong. As environments grow more complex, that gets harder to sustain. AI tools showing up on laptops and servers, often outside IT's view, only add to the pile.
Seeing isn't understanding, and understanding isn't acting. The ground stop showed how quickly one fault can become a nationwide disruption. The window to act is usually earlier, while the problem is still small.
That's the gap Autonomous Endpoint Management (AEM), which we're bringing to market, is built to close. AEM is an AI partner for endpoint operations. It watches what's happening across the estate, recognizes what matters, and recommends what to do. When an action is repeatable and well understood, it can carry it out within a defined playbook, and it keeps a record of everything it did. Its autonomy grows the way a new technician does: first it observes, then it assists under supervision, then it acts on its own for the work it has proven it can handle. People stay in control and decide how far that goes.
Two payoffs matter most for continuity.
Capacity. Continuity planning is important, but rarely urgent, so it loses to the ticket queue almost every time. Each routine task an AI partner takes on, such as patching, remediation, and the repetitive fixes that fill a week, returns time to the team. That time can go to the planning and exercises that keep getting postponed, but only if leaders point it there on purpose.
Change. Remember the third source of disruption: an untested change made by someone who couldn't see what it would break. Endpoints are where many changes land, from operating system updates to new applications and AI tools. Rolling changes out in stages, watching what happens, and keeping a way to roll back is degraded-mode thinking applied to everyday operations. Fail small, fail visibly, fail reversibly.
The goal isn't to take people out of the loop. It's to catch problems while they're still small, and to give people the time and the information to plan for the ones that still get through.
What to do next
Pick one service the business can't run without, such as payments, ordering, or sign-in.
Ask an AI partner to list what could disrupt it: natural, deliberate, and accidental.
With the people who depend on that service, rank the list by likelihood and impact.
Check how changes reach that service: who approves them, how they're staged, and how you'd roll one back.
For the top three risks, define a degraded mode: what keeps running, who decides, and how you'll communicate.
Run a tabletop exercise. Then repeat the cycle with the next service.
A loop, not a leap
The larger conversation we want to lead is about the nature of work itself: what work is worth doing, who or what should do it, and how people and AI partners share it. That's the future SuperOps is working toward: operations that understand what's happening, recognize what matters, and act early, with people in control.
Planning for disruption isn't a one-time project. Like the disaster cycle, it's a loop, and every pass makes you better at it. We're all early in this journey. This isn't a time to sit on the sidelines. It's a time to act, experiment, and learn what's possible now, so we know where to go next.
This is the slowest the pace of change will ever be. Don't wait for your own ground stop. Start the loop now.