At the recent VB Transform 2026 conference, Instacart Chief Technology Officer (CTO) Anirban Kundu surprised the audience by declaring that artificial intelligence (AI) has completely eliminated the company's technical debt worries. According to Kundu, the company's software engineers no longer need to read source code in 97% of cases, as the majority of repetitive coding tasks have been fully handed over to AI agents. This allows the engineering team to focus on core problems that require critical thinking and complex exception handling.
Key Details
The decision to offload coding to AI stemmed from the fact that developers often spent too much time on repetitive, tedious coding tasks. At Instacart, AI agents now handle most of the source code and boilerplate generation, particularly in new projects where code is regenerated weekly. When a piece of software is no longer actively used, the system automatically discards and rebuilds it from scratch instead of attempting to patch or maintain it. This eliminates the traditional concept of 'technical debt', similar to how compilers previously handled assembly code.
However, humans still play a decisive role in the remaining 3% of the work. This small percentage primarily lies in legacy systems, strictly compliant systems, or latency-sensitive pipelines. To fully address this, Instacart is deploying a project called 'Atoms' to break down monoliths into leaner, modular RPC-driven (remote procedure call) service architectures under human supervision.
Technical & Technological Analysis
This shift also redefines the software testing and review process at Instacart. Instead of conducting traditional manual code reviews—which are no longer necessary as AI consistently ensures correct syntax—the company is moving toward an 'intent model'. Engineers are trained to prompt and verify whether the AI correctly understands the business intent. The system now runs about 7,000 automated reviews per month while responding to over 8,000 real-time queries from developers with 99.9% accuracy.
In addition to writing code, Instacart has built an agentic Site Reliability Engineering (SRE) system. Unlike generic models on the market, this AI SRE was trained directly on years of Instacart's own incident data and root-cause analyses. Consequently, the detection and resolution rate for real-world operational incidents has surged from 60% to over 90%. An internal AI tool called 'Blueberry' monitors over 200 Slack channels and system alert signals simultaneously to detect anomalies that human intuition might miss.
Expert Perspectives & Insights
Kundu noted that human intuition can sometimes hinder incident resolution, as people tend to fall back on old cognitive biases before debugging. In contrast, the AI SRE system takes a more holistic view, making decisions based on the entire data landscape without bias. For instance, during a database shard fragmentation incident caused by an AWS EBS storage issue, while engineers were struggling to identify the cause, Blueberry took only 20 minutes to pinpoint the exact bottleneck in the company's internal feature-flag delivery system.
Impact & Future Outlook
In this new era, the role of software engineers at Instacart is being completely redefined. Instead of writing code line by line, their primary job will be guiding AI systems, designing evaluation pipelines, orchestrating parallel testing, and handling edge cases. Furthermore, Instacart is moving toward democratizing code by embedding expertise directly into technical specification documents that any team can use, rather than having a single siloed team own and modify the source code.