In the high-stakes world of peer-to-peer car sharing, where millions of travelers and hosts rely on a seamless digital experience, downtime is not merely an inconvenience—it is a direct threat to the business. Turo, the world’s largest marketplace for car sharing, operates a sophisticated, high-velocity engineering ecosystem defined by complex microservices and expansive Kubernetes clusters. Yet, as the company embraced the generative AI revolution, it hit a wall that many modern enterprises are currently facing: the "AI Paradox."
The paradox is simple yet debilitating: generative AI tools have mastered the art of writing code, but they remain profoundly ignorant of the production environments where that code lives. For Turo, this led to a surge in coding velocity that, ironically, slowed down the entire organization. In a recent fireside chat, New Relic’s VP of Developer Relations, Jemiah Sius, sat down with Turo’s VP of Engineering, Victor Mora, to dissect how the company overcame this bottleneck by integrating production-grade observability with autonomous AI agents.
The AI Paradox: Why Faster Coding Leads to Slower Pipelines
The software industry is currently undergoing a structural shift. According to New Relic’s 2026 State of AI Coding Report, a staggering 67% of technology leaders report that AI now generates or significantly refactors between 51% and 75% of their organization’s weekly code output. While this shift has undeniably accelerated the initial development phase, it has introduced massive inefficiencies downstream.
At Turo, the rapid adoption of local AI coding assistants meant that developers were churning out pull requests at an unprecedented rate. However, this high-volume output quickly overwhelmed the company’s code review and deployment pipelines.
"Writing code is the easy part," Victor Mora noted during the discussion. "The hardest part now is accurately defining the problem you’re trying to solve. We were writing so much code that our deployment pipelines—particularly for our legacy monoliths—became a significant productivity bottleneck."
The core issue lies in context. AI tools possess full visibility into source code syntax and local IDE environments, but they have zero visibility into real-world production traces, dependencies, or the intricate state of a live Kubernetes cluster. When an engineer uses AI to write a feature, the tool assumes a "happy path." When that code hits a complex, multi-layered production environment, it often creates unexpected ripples, leading to failed builds, security regressions, or performance degradation.
Chronology of a Crisis: From Toil to Automation
Before the integration of New Relic’s observability platform with their AI workflows, Turo’s on-call engineers were suffering from severe "alert fatigue." When a microservice failed, the engineer on duty was often tasked with triaging a service they did not build or maintain. This forced them to manually correlate logs, trace distributed requests across disparate clusters, and decipher error messages—all while the clock ticked toward a potential SLA breach.
The Midnight Incident
The true test of Turo’s new operational strategy arrived during a critical production incident late one night. Under the old paradigm, a midnight alert would have triggered a chaotic scramble:
- The Alert: An automated notification hits the engineer’s phone.
- The Wake-up: The engineer struggles to log in, access the VPN, and navigate a complex dashboard.
- The Hunt: The engineer manually searches through logs to identify the root cause of the service failure.
- The Mitigation: Once identified, they write a fix, run a test, and deploy.
With the new, AI-augmented workflow, the outcome was drastically different. The integration of New Relic with cloud-based AI agents allowed for a "60-second root cause" resolution.
Within the first minute of the alert firing, the following occurred:
- Automatic Contextualization: The AI agent, equipped with real-time telemetry from New Relic, immediately ingested the error logs and cross-referenced them with recent deployment metadata.
- Automated Triage: The agent identified that a specific microservice was failing due to a memory leak caused by a recent configuration change, rather than a code bug.
- Proactive Reporting: Before the engineer had even fully opened their laptop, the agent had already generated a summary of the incident, pinpointed the offending code commit, and suggested a rollback strategy.
The result? The incident was mitigated in minutes, not hours. The engineer’s role shifted from "firefighter" to "supervisor," approving the automated recommendations rather than performing manual labor.
Supporting Data: The Cost of Ignorance
The challenges faced by Turo are backed by industry-wide data. The 2026 State of AI Coding Report highlights that while developers are spending less time writing boilerplate, they are spending significantly more time managing "AI debt." This debt manifests in three primary areas:
- Deployment Bottlenecks: As AI generates more code, the surface area for code review increases, creating a backlog that stunts the velocity gains provided by the AI in the first place.
- MTTR (Mean Time to Resolution) Inflation: In the absence of production awareness, AI-generated code is often optimized for speed rather than stability, leading to more frequent, harder-to-diagnose production failures.
- Operational Fatigue: Engineering teams are forced to spend their "cognitive budget" on triaging alerts for microservices they lack deep familiarity with, leading to burnout and decreased innovation capacity.
Turo’s pivot to "Production-Grade Observability" aims to solve these by grounding AI in reality. By feeding live telemetry back into the AI agent, Turo ensures that the AI is not just a coding assistant, but an operational partner that understands the health of the entire system.
Official Perspective: Engineering for the Future
During the fireside chat, Victor Mora emphasized that the transition was not just about buying new tools, but about fundamentally changing the engineering culture at Turo.
"We had to move beyond local laptop coding," Mora explained. "By deploying context-aware AI agents directly into our production workflows, we gave our engineers the ‘sight’ they needed to manage our scale."
New Relic’s Jemiah Sius underscored that this shift is the new standard for high-performance teams. "You cannot control AI agents without complete observability," Sius remarked. "If you allow AI to write code, you must allow observability to validate it. The gap between the PR and the production reality is where most companies lose their competitive edge."
Strategic Implications: The Path Forward
For engineering leaders looking to emulate Turo’s success, the path forward requires a shift in how AI agents are deployed. It is no longer sufficient to treat AI as a standalone productivity hack. Instead, it must be integrated as a functional component of the observability stack.
Key Implications for Engineering Leadership:
- From Code-Centric to Context-Centric: Shift the focus from how much code is being generated to how much context is being applied to that code.
- Grounding AI in Data: Utilize tools like New Relic’s Ground Truth to provide an operational context layer. By equipping agents with native tools that allow them to query real-time telemetry, organizations can turn raw data into immediate, accurate action.
- Standardizing Agent Governance: Establish clear rules for when and how an AI agent can intervene in production. Turo’s success was predicated on having the AI act as an advisor, with the human engineer acting as the final arbiter of change.
- Prioritizing Developer Experience (DevEx): Use AI to solve the toil associated with on-call rotations, not just the toil associated with writing functions. When engineers feel supported by technology during a crisis, retention and morale improve.
Conclusion: The Future is Autonomous, but Grounded
Turo’s journey serves as a blueprint for the modern enterprise. As we push toward a future where autonomous agents manage increasingly complex microservices, the bridge between code and production must be fortified by data. By grounding AI in production-grade observability, Turo has not only reduced its MTTR and improved its system design, but it has also reclaimed the most valuable asset in the software industry: the focus of its engineers.
As the industry continues to evolve, the distinction between those who use AI to generate code and those who use AI to manage production will become the defining line between market leaders and those left behind. For Turo, the future of engineering is not just faster—it is smarter, safer, and inherently more resilient.
