Instacart CTO: AI adoption reduces need to manage technical debt
Instacart's Chief Technology Officer, Anirban Kundu, has suggested that the company is significantly reducing its concern over technical debt due to the adoption of AI agents. At the VB Transform 2026 event, Kundu explained that AI is now handling the majority of repetitive and high-volume tasks previously performed by human engineers. This allows Instacart's developers to focus on more complex problems requiring judgment, intent, and exception handling. Currently, human engineers at Instacart do not read code in 97% of cases, as AI agents generate and regenerate code, especially for newer projects. Kundu stated that this approach means technical debt is less of a concern because inactive code can simply be dropped and rebuilt. However, human oversight is still required for 3% of cases involving legacy systems, compliance, latency-sensitive workflows, or complex legacy codebases. Instacart is addressing these areas through a project called Atoms, which breaks down and rebuilds systems in a modular fashion. The company is also shifting towards remote procedure call (RPC)-driven architectures. A significant challenge remains in evaluating AI-generated code, leading Instacart to focus on an "intent model" where developers are trained to ask AI the right questions. The company conducts approximately 7,000 automatic evaluations monthly and handles over 8,000 real-time developer queries with high accuracy. Furthermore, Instacart has developed an agentic site reliability engineering (SRE) system, trained on its own incident data, which has improved the detection and mitigation of production issues from around 60% to over 90%. This AI SRE, named Blueberry, monitors numerous communication channels and system signals, identifying subtle issues that human intuition might miss. For example, Blueberry quickly identified a database shard problem linked to an inadequate feature rollout, resolving the incident faster than human teams might have. Kundu envisions the future role of engineers to involve navigating AI systems, designing evaluation processes, managing experiments, and identifying edge cases, rather than solely writing code.
Instacart's embrace of AI for code generation and system management represents a significant shift in software development paradigms. By offloading routine coding and repetitive tasks to AI agents, the company aims to mitigate the traditional burden of technical debt, which often arises from quick fixes and evolving requirements. This strategy allows human engineers to concentrate on higher-level problem-solving and system design, potentially accelerating innovation. However, the reliance on AI for code generation introduces new challenges, particularly in robust evaluation and ensuring the integrity of complex, legacy, or mission-critical systems. The company's focus on an "intent model" and specialized AI SRE systems suggests a forward-thinking approach to managing these new complexities. As AI capabilities grow, organizations will need to continuously adapt their governance and oversight frameworks to balance efficiency gains with inherent risks, ensuring that human judgment remains central to critical decision-making and system resilience in the long term.
AI-generated to prompt reflection — not editorial opinion, not advice, not a statement of fact. How this works.