Generative AI is reshaping industries, and the world of technology engineering is no exception. From breathless headlines about AI writing the majority of code to the practical realities on the ground, separating hype from impact is a constant challenge. To get a clear picture, we sat down with Andrea Reed, Head of Technology Engineering at Convex, on our Unstructured Unlocked podcast.
Andrea shared her firsthand experience with integrating generative AI into engineering workflows, the real productivity gains, the persistent challenges, and her vision for the future. This post unpacks her key insights, offering a grounded perspective on how AI is truly influencing software development, data strategy, and the future of engineering careers in the complex world of insurance.
The Reality of AI-Supercharged Engineering
While some reports suggest a complete takeover by AI code generators, Andreaās experience at Convex paints a more nuanced picture. The pressure from the industry and investors, fueled by bold claims from tech giants, is undeniable. However, the practical application of these tools reveals a collaborative relationship between human developers and AI, rather than a replacement.
Related content: Innovation vs. Risk in Insurance Decisioning: Leveraging AI to Transform Strategy
Productivity: A Boost, Not a Revolution
At Convex, tools like GitHub Copilot are actively used, and Andrea estimates that about 25% of their code is AI-generated. This is a significant boost to productivity, but it comes with a critical caveat: every line of AI-generated code requires careful review and refinement by an expert developer.
These tools are particularly effective at accelerating more straightforward tasks and helping junior developers get started. Senior engineers can offload routine coding, freeing them up to focus on more complex architectural challenges. While productivity has certainly increased over the last year, Andrea is clear that the technology has not yet reached a point where developers can be swapped out for AI agents.
The Context Challenge in Complex Architectures
One of the primary hurdles for AI is understanding the full context of a complex IT environment. While large language models now have bigger context windows, allowing them to analyze an entire codebase, they still struggle with the bigger picture.
Convex operates on a microservices architecture, where numerous distinct components and integrations must work together seamlessly. According to Andrea, AI has difficulty grasping this overarching structure and the intricate relationships between different services. This lack of architectural awareness is a major reason why human oversight remains indispensable. The system’s ability to pattern-match is strong, but it falters when faced with high variability and architectural complexity that require genuine understanding.
Enhancing Data Strategy with Unstructured Data
The insurance industry runs on data, but much of its most valuable information is locked away in unstructured formats like documents, emails, and reports. Andrea highlighted how generative AI is poised to unlock this potential and enhance their existing data strategy.
The goal is to move beyond structured data and incorporate the vast reserves of unstructured information. By vectorizing documents, teams can perform semantic searches and ask complex questions across their entire document repository. This allows them to surface insights that were previously inaccessible. For Convex, the immediate focus is on integrating this newly available data to enrich their decision-making processes, marking a key strategic priority for the coming months.
Related content: Beyond Document Extraction: How AI Is Reshaping Insurance Decisioning
The Human in the Loop: Managing a Future of AI Agents
As the industry moves toward more sophisticated “agentic AI”āwhere AI agents can perform tasks autonomouslyāthe conversation shifts to governance, compliance, and management. Andrea emphasizes that for now, and for the foreseeable future, the “human in the loop” is not just a concept but a necessity.
Building confidence in AI’s outputs is a gradual process. In a highly regulated field like insurance, ensuring compliance and accuracy is non-negotiable. Confidence levels in data extraction and AI-generated answers must be meticulously tracked and validated. Just as you wouldn’t hire a new employee and check on them six months later, AI agents require constant monitoring, especially in their early stages.
The future role of a technical leader might evolve into a “manager of agents,” responsible for their performance, behavior, and accuracy. This involves setting up standardized telemetry to monitor all agents, identify misbehaving ones, and ensure explainability for regulators who may demand audits of past transactions.
Advice for the Next Generation of Engineers
With headlines proclaiming the “end of the software engineer,” what should aspiring technologists do? Andrea, whose own daughter is studying computer science, offers clear and practical advice that cuts through the noise.
She refutes the idea that coding is becoming obsolete. While the nature of the job will change, a fundamental understanding of programming remains crucial. She advises young engineers to learn how programming languages work, understand different paradigms, and grasp the principles of generative AI.
More importantly, she stresses the need for a solid grasp of the basics: operating systems, computer architecture, and troubleshooting. While AI can handle many tasks, it can also fail. When it does, it will be the engineers who understand the underlying mechanics of computing who will be able to solve the problem. The engineers of the future will need to combine this foundational knowledge with the ability to control and manage AI systems effectively.
What’s Next? Scale, Security, and Beyond
When asked what the conversation will be about in a year, Andrea pointed to two key areas:
- Scale: While Convex is currently focused on identifying the right use cases and implementing agents effectively, the challenge of scaling these solutions across the enterprise will soon become a primary concern. How do you successfully expand AI initiatives across different lines of business and handle fluctuating volumes, like the annual renewal season?
- Security: As the adoption of generative AI grows, security will become an even more critical topic. Protecting proprietary data used to train models and securing the outputs of AI agents will present significant challenges that the industry must address.
Andrea Reedās insights from the front lines of technology engineering provide a valuable, pragmatic view of generative AI’s role today. It is not a magical solution but a powerful tool that, when wielded by skilled professionals, can drive significant productivity and unlock new opportunities. The future is not one of human versus machine, but of a powerful partnership that elevates what’s possible.
Subscribe to the Unstructured Unlocked podcast to get the latest episodes on your favorite platforms, including: Apple PodcastsSpotifyAmazon Music
