Why only 37% of developers trust AI for incident response

IT uptime is now firmly established as a board-level priority. It’s not hard to see why organizations are paying a heavy price for digital disruptions. A recent PagerDuty study reveals that 68% of global organizations lose more than $300,000 per hour during IT incidents.

This isn’t just bad for customer loyalty and the bottom line. More than two-fifths (42%) of operations leaders recognize that outages can also contribute to developer burnout. But by effectively combining human teams and autonomous AI tools, organizations can reduce the workload on responders, which can, in turn, help improve resilience, operational maturity, and innovation.

Where does fatigue come from?

Developer fatigue is the price global businesses are paying for the seamless, digital-first experiences they promise customers. Engineering teams are under constant pressure to deliver updates, but the complex, brittle architectures they work with make incidents inevitable. Each new release brings new dependencies and potential sources of failure.

When something goes wrong, engineering talent is pulled away from high-value work to investigate and remediate. It doesn’t help that the siloed tooling they are stuck using can’t cancel alert noise, intelligently route incidents, or proactively find and fix well-understood issues. The average team might be forced to handle thousands of daily alerts, many of which are irrelevant or duplicated.

“The average team might be forced to handle thousands of daily alerts, many of which are irrelevant or duplicated.”

Automation can help, but it can also perpetuate organizational silos, produce misleading results, and create more work if teams are needed to debug scripts and manage exceptions. The result is sinking morale as exhaustion sets in, and firefighting trumps innovation.

There are ways to avoid this fate. AI offers potentially big gains for organizations that work out how to harness it in a developer-aligned manner. 

Consider the following three steps:

1. Build trust among developers

Without trust in AI, there can be no progress towards a more mature and developer-friendly model. Confidence in the technology is high among IT and business leaders. Research indicates that 59% of IT decision-makers expect AI to improve downtime and recovery performance by more than 20%. However, developers appear not to share the same enthusiasm, with only 37% agreeing. Changing these perceptions is therefore a critical first step to reducing burnout.

“Without trust in AI, there can be no progress towards a more mature and developer-friendly model.”

Organizations will need to evangelize and help teams understand how AI can help them personally and professionally. The key is showing how it can eliminate repetitive toil and free them up to work on more creative and rewarding tasks.

However, no two team members are the same. Some will require more convincing than others because they’re unsure of AI’s value or worried about job security, or both. It’s important to lead with empathy. Leaders will know these efforts are working if team members start turning up to optional workshops, trying new tools, and sharing learnings without being asked.

2. Upskill existing teams

Arguably, the most challenging part of the project comes next: empowering developer teams with skills to use AI tools in their daily work. Around half (51%) of global organizations are expecting to hire or reskill to deliver AI-powered incident detection and response.

The key isn’t to enroll them in generic training courses, which will do little more than impart surface-level knowledge. Instead, upskilling must be relevant to each individual and include the tools they use, the languages they code in, and the workflows they operate with.

Begin with the core skills required to build AI fluency. This could include problem decomposition, the breaking down of tasks into individual components that AI can handle. Or prompt engineering to communicate effectively with AI systems and understand how different models interpret inputs. Developers also need a grounding in quality evaluation, enabling them to effectively determine when AI output is good enough and when human input is required.

Next, customize curricula to make hands-on courses relevant to specific roles. Use the same tools, data, and workflows to make courses as contextually relevant as possible. Remember that this all takes time, and training should be considered a continuous program.

3. Combine humans and AI agents effectively

As important as AI is in reducing the workload of developer teams, it’s not a panacea. Caution remains the default setting for many organizations, especially when appraising autonomous, agentic AI systems. Nearly two-thirds (62%) are looking for an even mix of human and agentic AI work over the coming three years. This makes it vital to clearly define when a human in the loop is required.

One way to do this is to follow a three-tier model. 

The first tranche consists of routine issues with known fixes. AI handles all of these, from detection to remediation. Humans are only required to review reports and refine processes post-event to learn and improve. At the other end of the scale are novel or complex issues that require human expertise and creative problem-solving. Humans will always take the lead here, using AI only to gather context, collect data, and handle routine communications.

In between the two extremes are familiar incidents with elements of ambiguity. In this tier, AI can be let loose on detailed analysis, data correlation, and generating recommendations. However, humans will make the final decision on remediation, ultimately becoming AI overseers.

Expanding the team

To get these efforts up and running, organizations need to work through several stages. 

First comes governance, then user training, and defining when to use humans in the loop. Next, organizations need to put in place guardrails to limit agent autonomy and mechanisms to log decisions, create audit trails, and improve transparency. 

Finally, they must track multiple metrics to monitor and continuously improve AI performance. An added benefit of this framework is that it helps to establish the engineers’ trust in the AI tools.

AI agents should be considered as new members of the team, so their impact must be evaluated just like that of their human counterparts. If these efforts are successful, organizations will benefit from happier teams, less burnout, and more resilient digital operations.