Instacart CTO: AI Has Reduced Company's Tech Debt

Instacart's Chief Technology Officer Anirban Kundu has stated that artificial intelligence (AI) can significantly reduce the company's technical debt by absorbing repetitive, high-volume software development tasks. This allows human engineers to focus on problems requiring judgment, intent, and exception handling.

Speaking at VB Transform 2026, Kundu explained that in 97% of cases, Instacart's developers no longer read code. A substantial amount of code generation and boilerplate work is handled by AI agents, particularly in new projects where code is regenerated weekly. "The benefit of that is we don't care about tech debt anymore," Kundu said. "Things that are not active just get dropped out and then it gets rebuilt."

Human oversight remains necessary for legacy, compliance, and latency-sensitive systems, as well as areas with significant amounts of dormant or partially active code. Instacart is addressing these through its project Atoms, breaking down and rebuilding systems in a cleaner, more modular form, transitioning from monoliths to RPC-driven architectures.

A key challenge lies in evaluation. With AI generating code, traditional code reviews become less relevant. The focus is shifting to an "intent model," where developers learn to ask the right questions of various models. Additionally, the company has developed an agentic site reliability engineering (SRE) system trained on Instacart's specific incident data and root-cause analyses, significantly improving the accuracy of production issue detection and mitigation.

Looking ahead, engineers' most tactical work will involve navigating AI systems, designing and supervising evaluation processes, and determining when to escalate issues. Domain expertise is also being redefined in the AI era, with Instacart aiming to democratize code management so that core knowledge is accessible across all teams.