Instacart’s CTO Claims AI Tech Debt Is No Longer a Concern

Instacart’s CTO Anirban Kundu told VB Transform 2026 that the company has turned the problem of AI tech debt into a non‑issue. He notes that ↑ 97% of code modifications are now generated by AI agents, freeing engineers to concentrate on judgment‑heavy tasks.

How AI Tech Debt Vanishes at Instacart

The new workflow treats repetitive, high‑volume chores as machine‑handled, while humans address intent, edge cases, and exception handling. In practice, developers rarely open a file; AI writes boilerplate and refactors stale sections automatically.

“The lines of code are correct, the syntax meets expectations,” Kundu said.

Legacy components still account for ↓ 3% of the codebase, demanding careful human oversight. Instacart’s Atoms initiative breaks monoliths into RPC‑driven modules, rebuilding them in a cleaner, more modular fashion.

Evaluation is now a separate layer. Roughly 7,000 automatic checks run each month, answering over 8,000 developer queries with 99.9% accuracy. The goal is an “intent model” where engineers ask the right questions of the AI.

On the reliability front, the firm deployed an agentic SRE system trained on its own incident logs. The AI monitors dozens of Slack channels, spotting patterns humans might miss. In one case, it flagged a database shard issue within minutes, linking it to a misbehaving feature flag, accelerating resolution.

Industry observers note that Instacart’s approach mirrors broader trends highlighted by Reuters, where AI‑driven development is reshaping software maintenance.

Looking ahead, engineers will spend most of their time steering AI outputs, designing evaluation pipelines, and coordinating experiments. Domain knowledge is being codified into shared specifications, allowing any team to modify code without a single “owner” bottleneck.