Phil Bliss/theispot.com
A vice president of product opens her laptop on a Monday morning to find that the AI model her team had worked with for the past six weeks to build a customer workflow has been leapfrogged by a cheaper, faster alternative. Again. Her Slack feed is blowing up with links to the announcement. The CEO has already forwarded an article about what a competitor is doing with the new tool, with the subject line “FYI.” She hasn’t finished rolling out the last integration, and now she’s wondering whether to scrap it. She is not resistant to AI. She is worn out by it.
Most leaders look at this picture and see an execution problem: The organization wasn’t moving fast enough. The cautionary tale that reinforces that instinct is Chegg, the education company whose market capitalization collapsed when the launch of AI-powered alternatives rendered its core tutoring model obsolete. The lesson everyone has drawn is obvious: Move fast or die. So leaders push harder, with more pilots, more mandates, and a constant drumbeat of urgency.
But that lesson, taken too literally, backfires. Bracing only against the danger of moving too slowly, leaders managing AI adoption underestimate a quieter risk: that they will wear out their organizations by racing toward a finish line that does not exist. The old playbook was built for disruptions that end, and its instincts (move faster, push harder, wait for things to settle) become liabilities when there is no end state. Leaders who optimize for speed alone will lose to those who build for endurance as well.
What follows is a reframing and a set of emerging practices for leading through an AI disruption that will not settle.
From Process to Permanent Condition
Research on disruption has been circling this problem for years. One influential strand that one of us (Rory) developed with Clayton Christensen and Michael Raynor pressed on the point that disruption is a process, not an event. The recurring incumbent error is to judge the threat by where it stands rather than where it is heading. Yet, even correcting for this carries a quiet assumption that the threat’s trajectory has an ultimate destination. After Netflix disrupted Blockbuster, streaming became the new normal. Each wave of new technology ran turbulently for a while and then hardened into arrangements a company could see and plan around.
AI changes this dynamic. With previous technologies, the entrant’s advantage grew because something outside it improved: Components got cheaper, networks got faster, supply chains got better. AI is increasingly self-improving. Each generation helps train and build the next, so the distance between waves keeps shrinking. There is no settled position to plan toward, because the core keeps extending and the old barriers to disruption keep falling. We have come to call this condition steady-state disruption: a context in which capability shifts arrive continuously and accelerate one another, with no equilibrium in sight.
Managers and scholars already have language to describe turbulent environments. They talk about VUCA (volatility, uncertainty, complexity, and ambiguity) and about the dynamic capabilities a company needs to sense change and adapt. But that vocabulary assumes that the turbulence eventually breaks: A period of upheaval is followed by a return to relative calm. Steady-state disruption is the condition in which the calm never comes.
If disruption is a process rather than a sequence of separate shocks, then the AI capabilities landing inside an organization are not a series of discrete events to be handled one at a time but an ongoing process. Companies that treat a continuous process as a string of episodes fatigue their employees and ultimately struggle to adapt.
The pace itself shows no sign of letting up. Stanford’s 2026 AI Index shows AI systems posting steadily higher scores on standard industry benchmarks. New frontier models appear every few months, and an industry investor reported that the leading model often holds its position for only a few weeks before a newer one or an open-source rival takes share.
This cadence is especially hard to absorb because the work never reaches a stopping point. As Airtable CEO Howie Liu has observed, AI adoption is unlike the move from desktop to mobile or from on-premises to cloud computing. Each of those shifts was a single, fairly foreseeable change in form, but with AI, every model release brings new capabilities and new patterns that have to be learned more or less from scratch. Even if this progress were to hit a sudden plateau, organizations would still need to spend years folding existing capabilities into their products, workflows, and decision-making.
Addressing the Human Cost
For a lot of people, the early excitement has curdled into something that’s harder to sustain. Recent research makes the cost concrete. One analysis found that AI tends to intensify rather than lighten individual workloads, piling on cognitive demand faster than it strips away drudgery. And Deloitte’s Global Human Capital Trends survey saw the same thing at the organization level. As collaboration with AI deepens, so, too, do burnout, loneliness, and overload.
These findings document strain on individuals, but the cause sits above the individual level. When a playbook written for episodic disruption no longer works, the organizational machinery that once absorbed shocks now transmits them directly to workers instead.
So, what should a change management toolkit for the age of continual AI disruption look like? A handful of practices that are provisional but useful are taking shape at leading companies. Instead of placing the burden of absorbing change onto individual employees, these practices move some portion of that burden onto the structure of the organization.
Practice 1: Build a Permanent AI Unit
Since 2022, a common response to the rapid rise of AI has been to form an AI committee — a group of people asked to advise, set direction, and evangelize on AI, usually on top of their existing jobs. Committees of this kind tend to add work rather than soak it up. Members are stretched thin, the rest of the organization gets only intermittent guidance, and the committee’s own queue keeps growing. Steady-state disruption calls for a sturdier, more permanent approach. Whether it is a full-time team at a large company or a carved-out slice of a few people’s time at a smaller one, the work of tracking, translating, and triaging AI’s churn should be somebody’s actual job rather than a standing favor.
Microsoft offers an illustration. The AI Center of Excellence inside Microsoft Digital began as an ordinary advisory group in 2023. But the group’s leader, Qingsu Wu, recalled that as adoption spread, so did duplicated effort, uneven governance, and gaps between strategy and implementation. The question, Wu said, shifted from “How do we help teams try AI?” to “How do we turn AI into consistent, measurable outcomes at scale?” The center became the place where AI work is coordinated, with a single idea intake pipeline, a hub for upstream architecture and security decisions, and the ability to see patterns where individuals and siloed teams cannot.
Once such a group has enough depth, the frontier-scanning activities and scrap-or-scale calls that used to land on scattered individuals become the standing remit of people equipped to handle them.
Practice 2: Run Two Clocks, Not One
Most organizations keep time on a single clock. Plans are quarterly, budgets are annual, and the all-hands meeting lands on its dependable schedule. Onto that steady rhythm, employees are now also being asked to ship AI experiments by the week, keep last quarter’s integrations running, follow a frontier that shifts constantly, and explain to leadership what any of it means for the business. Companies have always lived with some gap between fast work and slow work. AI has widened it past the point where one person can comfortably hold both ends.
The strain shows up first in product organizations, where the distance between weekly model releases and quarterly road maps is hardest to ignore. Airtable CEO Liu watched AI-native competitors shipping major capabilities every week while his own teams followed quarterly road maps. So he split the product organization into two groups, borrowing a distinction from psychologist Daniel Kahneman: A fast-thinking group ships AI capabilities on a near-weekly basis, while a slow-thinking group takes on the deliberate infrastructure bets — the kind of work that, as Liu put it, you cannot ship in a week via a “hacky prototype.” The two are meant to feed each other. The fast group surfaces new possibilities, and the slow group turns the promising ones into things the company can rely on. Without such a deliberate split to protect the slow clock, the faster clock becomes the standard against which everyone is measured.
Practice 3: Teach in the Flow of Work
Under constant change, most corporate training approaches, such as annual certifications or one-off workshops, fall behind almost as soon as they are delivered. Employees are left to keep pace with new developments on their own and end up concluding that the frontier is simply outrunning them. The feeling only deepens when the learning is stacked on top of a job that is already full.
Learning should instead be made continuous and specific to the role, delivered inside the work itself. There is good evidence that knowledge sticks better through short, repeated exposures spread over time than through one-off intensive sessions. Salesforce adopts this approach with its AI-powered internal platform, Career Connect, which reviews an employee’s existing skills, identifies the gaps between those skills and their aspirations, and serves up tailored learning opportunities (such as courses, stretch assignments, or mentorship) through Slack, where they already work. The organization’s Agentforce Learning Days add a recurring, companywide push on AI skill development specifically. What emerges is an architecture for AI fluency and ongoing development that is steadier and closer to the work, better aligning with the realities of steady-state disruption.
What complicates this approach when it comes to AI is that employees who fear being replaced by the technology have little incentive to engage with it seriously. CEO Greg Case at insurance and risk advisory firm Aon has addressed this fear directly. His bet is that AI will widen what the firm’s roughly 60,000 employees can do rather than substitute for them. He has built Aon’s investment in continuous AI fluency around that framing and has credibility with his employees for leading Aon through the pandemic without layoffs. Continuous learning requires continuous buy-in, and buy-in requires workers to believe that getting better at AI benefits them, not just the organization.
When training stops being a place employees go and becomes part of how they work — small and constant rather than disruptive and periodic — it keeps a workforce current without asking people to absorb the frontier on their own time.
The vice president of product we mentioned at the beginning was faced with a new-model announcement, a forwarded article in her inbox, and a newly built workflow that was already obsolete. This steady-state disruption is exhausting because the weight of all this change rests on her and her coworkers, with nothing in the organization’s design built to help them carry it.
A generation of managers learned that disruption was a phenomenon that eventually settled. AI shows no sign of settling. Leaders who keep treating it as a series of episodes will keep piling that weight onto their people. Those who build organizations designed to carry it, through permanent AI infrastructure, split cadences, and work that has learning embedded into it, will be the ones who endure.
