A year ago, AI coding tools worked alongside developers—suggesting the next line, completing a function, accelerating work that a human was already doing. Today, they’re writing entire modules and services independently, producing code that no human has reviewed line by line, built from components that no single person has fully mapped. And adoption is only accelerating: According to our recent AI Resilience Survey, 84% of organizations are now using AI to write, review, or suggest code.

The volume and velocity of deployment have grown, and so has the complexity of what operations teams must manage. More components deployed mean more interactions, and more interactions mean more ways for things to break. 

Our 2026 State of AI-First Operations report, which surveyed 1,000 business and IT decision-makers and senior developers across seven global markets, found that more components, more interactions, and more AI-generated code mean more opportunities for something to go wrong—and the financial consequences when it does are significant. Sixty-eight percent of organizations lose more than $300,000 per hour during a major incident. For organizations investing in AI-driven development without investing equally in operations, that number reflects a risk that compounds with each deployment. How quickly you recover from failure is becoming as strategically important as how quickly you can ship.

The math on complexity

Software has always been built from components—modules, services, dependencies—that interact in ways that become harder to untangle as the system grows. A system built from three components has a manageable number of potential failure points. A system built from 30 has an order of magnitude more. At 300, the number of possible failure points multiplies faster than any team can track. And amid rapid AI adoption, organizations are struggling to keep up.

According to our AI Resilience Survey, 76% of organizations that have deployed AI agents already believe AI-driven complexity will soon outpace the number of people available to manage it, and 85% say they need better procedures to detect AI errors and failures before they cause disruption.

In the context of AI-powered development, for example, among the hardest errors to catch didn’t exist before: hallucinations in code. An AI agent can build an entire service that references dependencies that don’t exist—plausible-looking code that compiles, passes initial tests, and breaks in production in ways that are difficult to diagnose.

When firefighting eats innovation

AI-accelerated development promises more shipping velocity and more engineering time for work that actually moves the business forward. But that promise doesn’t always pan out. More code means more incidents. More incidents mean more engineers pulled away from building to fix things. Deploy, break something, fix it, deploy again, break three more things. 

According to our 2026 research, 45% of organizations expect AI-powered operations to deliver more time for teams to innovate and experiment. Yet incident response still consumes at least half of the time for 73% of developers, suggesting that resilience gains from AI have not yet translated into more time for innovation.

Every hour spent on incident response is an hour not spent on the work AI was supposed to unlock. When major incidents hit, organizations report consequences that extend well beyond lost revenue—brand damage (52%), recovery costs (50%), lost productivity (48%), and developer burnout (42%) all follow.

Recovery as a competitive advantage

According to our 2026 report, 95% of organizations say their leadership recognizes that faster recovery creates a competitive advantage. The organizations acting on that belief are pulling ahead: 82% of revenue-growing organizations are increasing operational resilience budgets, compared with 62% of those with flat or declining revenue. That investment is showing up in outcomes: 74% of revenue-growing organizations report improved resilience over the past year, compared to 64% of those with flat or declining performance.

Paradoxically, the organizations seeing the strongest resilience gains are also the ones leaning furthest into AI—not just for development, but for operations. Of the organizations that report improved resilience over the past year, those already using AI in their operations are seeing stronger gains: 75% versus 66% among those not yet using it.

Turning incidents into intelligence

Every incident contains information: what broke, why, how the team responded, and what worked. Organizations that capture and apply that intelligence systematically build operations that improve with every cycle—faster detection, better response, fewer recurrences. The companies that don’t resolve the incident and move on are starting from scratch each time.

The gap between the two is significant. Every organization we surveyed agrees that post-incident learning must be strengthened, yet only 48% currently turn incidents into structured learning opportunities. As AI-driven development introduces more complexity and more failure surface area, that gap will only widen.

Building operations that can keep pace

The volume and complexity of what engineering teams are deploying will continue to grow. Operations that can’t absorb that growth become the bottleneck.

PagerDuty is built to help organizations manage that complexity—introducing AI across the full incident lifecycle, from detection and triage through response, resolution, and post-incident learning, so that every incident makes the system smarter rather than just getting resolved and forgotten.

For a deeper look at how the most resilient organizations are approaching that work, and where the gaps are widening, read the full PagerDuty 2026 State of AI-Powered Operations report.