When Failure Isn't an Option: Building AI for the Real World

Building Trust in Autonomous Systems

When artificial intelligence moves beyond the digital realm into the physical world, the consequences of failure become tangible. An AI chatbot might give a wrong answer; an autonomous aircraft or vehicle, however, can ground a plane or cause a crash. The question isn't just about accuracy anymore—it's about trustworthiness.

This fundamental shift in perspective is what drives the conversation at TechCrunch Disrupt 2026 on the Real World AI Stage. Nathan Michael from Shield AI, Raquel Urtasun from Waabi, and Mikell Taylor from General Motors will share their insights into how they navigate the complexities of deploying AI systems where failure has real-world consequences.

"When failure isn't an option, 'almost ready' isn't enough," says Daniel Carter, Senior Business Correspondent at Newsclip. "We need to understand what it truly means for an autonomous system to be trustworthy in high-stakes environments."

Shield AI: Mission-Critical Autonomy

Nathan Michael's journey into autonomous systems began at Carnegie Mellon University, where he directed the Resilient Intelligent Systems Lab. His expertise spans AI, control systems, perception, and multi-robot coordination—fields that have culminated in Shield AI's platform-agnostic mission autonomy software, Hivemind.

Hivemind has already proven its worth in one of the most challenging environments: military operations. In February, the system was selected as an autonomy provider for the U.S. Air Force's Collaborative Combat Aircraft drone prototype program—a testament to the robustness and reliability that Michael and his team have built into their platform.

The company's recent $1.5 billion Series G funding at a $12.7 billion post-money valuation speaks volumes about investor confidence in their approach. But beyond financial metrics, what makes Shield AI stand out is how they balance performance with assurance—a critical distinction in high-stakes autonomous operations.

Michael's perspective brings to light the complexity of ensuring mission-critical autonomy. When systems are deployed in contested environments or under pressure, they must not only perform but also be validated and trusted by operators who depend on them. This isn't just about software—it's about safety culture, testing protocols, and regulatory navigation.

Waabi: The Road Ahead for Autonomous Driving

Raquel Urtasun's journey in autonomous vehicles spans over two decades. Before founding Waabi, she served as chief scientist at Uber ATG and continues to lead research at the University of Toronto. Her extensive academic and industry background positions her perfectly to tackle one of the most pressing questions in autonomous driving: How do we know when a system is truly ready for deployment?

Waabi's approach to autonomous driving emphasizes rigorous testing and validation through its Waabi World simulator, which trains and stress-tests the Waabi Driver in virtual environments. Despite recent funding and partnerships with Uber to support the deployment of 25,000 or more robotaxis, Urtasun has made it clear that full validation is still needed before driverless operation begins.

This focus on comprehensive testing isn't just a technical necessity—it's a safety imperative. The transition from simulation to real-world driving involves understanding edge cases, regulatory compliance, and public trust. Urtasun's experience in both research and industry provides valuable insight into how leaders can balance innovation with responsibility.

General Motors: Robots That Work Alongside People

Mikell Taylor brings a unique perspective to the discussion with her extensive background in robotics. From building a robotic senior prom date to leading Amazon Robotics' Proteus project, Taylor's career has always focused on practical, reliable systems that work effectively alongside humans.

As the Director of Robotics Strategy for General Motors' Autonomous Robotics Center, Taylor's focus lies on industrial applications where robots must be dependable and adaptable in dynamic environments. Her experience working with both autonomous underwater vehicles and industrial robotic systems gives her a deep understanding of how to design systems that people can trust.

The human factor is crucial when AI-powered machines enter the real world. Unlike laboratory settings, these systems operate around people who must rely on them for productivity and safety. This means user experience, adoption strategies, and continuous improvement must be built into every phase of development.

Comparing Approaches Across Domains

The session at Disrupt 2026 brings together three very different but equally critical approaches to autonomous AI development:

• Military/Aerospace (Shield AI): Emphasis on mission assurance, resilience, and interoperability across platforms
• Transportation (Waabi): Focus on safety validation, regulatory readiness, and public trust in autonomous vehicles
• Industrial Robotics (General Motors): Emphasis on practicality, human-robot collaboration, and real-world deployment success

Each approach addresses the same fundamental challenge: how to ensure that when systems leave controlled environments, they do so with confidence. The differences lie in how they define and measure readiness.

The Cost of Not Being Ready

When failure is not an option, the cost of inadequate preparation becomes extremely high. Consider:

• Financial Impact: A single autonomous vehicle accident can cost millions in liability, repairs, and lost productivity
• Public Trust: High-profile failures can undermine public confidence in AI systems, setting back innovation efforts
• Regulatory Consequences: In heavily regulated industries like aviation or automotive, compliance issues can halt development entirely

Michael, Urtasun, and Taylor all emphasize that trust isn't built overnight. It's earned through consistent performance, transparent testing, and accountability when things go wrong. These leaders understand that building systems that are ready for deployment means investing in the processes, culture, and infrastructure that support long-term success.

What's Next for Autonomous AI?

The conversation at Disrupt isn't just about what has been accomplished—it's also about where we're headed. As AI continues to evolve and integrate into more aspects of our lives, the question of readiness becomes even more complex.

We're seeing increased investment in safety protocols, validation frameworks, and human-AI interaction models. The key is to move beyond just technical capability toward systems that are fundamentally designed with trust at their core.

Whether it's in defense, transportation, or industrial applications, the goal remains the same: create AI systems that not only function but inspire confidence. In an era where artificial intelligence increasingly shapes our physical world, this trust isn't just desirable—it's essential.

At TechCrunch Disrupt 2026, these leaders will offer their perspectives on navigating the critical path from innovation to deployment. Their insights will be invaluable for anyone working in or interested in autonomous AI development—particularly as the industry continues to push boundaries and redefine what's possible.

Looking Forward

The challenges these three leaders face represent broader trends in AI development. As we move toward more autonomous systems, the need for robust testing, clear validation standards, and responsible deployment becomes paramount.

This session at Disrupt offers a rare opportunity to understand how industry pioneers approach the difficult question of readiness when failure isn't an option. The insights shared will help shape not just individual companies' strategies, but the entire ecosystem's approach to developing trustworthy AI systems.

For those who missed this discussion or want to revisit the conversation, the session serves as a reminder that building reliable AI systems is about more than technology—it's about responsibility, transparency, and ultimately, trust.