Quality assurance · Production

AI-Driven Development at Instacart: Scaling Impact and Increasing Velocity

The problem

Instacart's AI adoption began as informal, grassroots experimentation by individual engineers. To make adoption consistent and durable at organizational scale, they needed a structured approach beyond organic Slack threads and demos.

First attempt

AI coding agents struggled with large legacy monolith codebases where unclear boundaries led to slower suggestions, incomplete answers, or hallucinated code. Very large files (5,000+ lines) caused repeated task failures, and AI code translation sometimes carried over flawed logic without correcting it.

Workflow diagram · grounded in source
1
AI tools embedded in daily tasks
trigger
“engineers embedded AI tools like Ava, Cursor, and Glean directly into their day-to-day tasks”
2
AI agent code cleanup
ai_action
“the AI agent not only identified unused paths but rewrote the class with precision. We cleaned up 15+ feature flags from a single service just with the use of an AI coding agent in a single PR”
3
Figma-to-scaffold generation
ai_action
“engineers uploaded Figma screenshots of UI components. The AI interpreted layouts and generated usable scaffolds, turning static mocks into functional components with impressive fidelity”
4
AI-assisted debugging
ai_action
“we used our AI coding agent to identify why certain error codes weren't returned from some cross-service API calls. The AI pinpointed that enabling a specific parsing option in one of our libraries will resolve the issue”
5
RCA documentation generation
output
“teams were able to rapidly analyze logs and stack traces, draft root cause narratives, and even auto-generate "Five Why's" to feed into RCA documentation”
6
Prompt engineering playbooks
feedback_loop
“This resulted in an organic sharing of techniques across teams and even led to the creation of internal prompt engineering playbooks”
Reported outcome

Instacart achieved up to 20% time savings on frontend workflows for Fizz, shipped Fizz from concept to customer-ready in a few short months, and cleaned up 15+ feature flags from a single service in a single PR using an AI coding agent.
Tasks that traditionally took several days were completed in hours.

Reported metrics
Time savings on frontend workflowsup to 20%
feature flags cleaned up in a single PR15+
Fizz time to marketconcept to customer-ready in just a few short months
task duration for iOS search featuretasks that traditionally took several days were completed in hours
Reported stack
AvaCursorGleanClaude-3.7Claude-3.5Figma
Source
https://tech.instacart.com/ai-driven-development-at-instacart-scaling-impact-and-increasing-velocity-43f6b3902a32
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Instacart achieved up to 20% time savings on frontend workflows for Fizz, shipped Fizz from concept to customer-ready in a few short months, and cleaned up 15+ feature flags from a single service in a single PR using…

What tools did this team use?

Ava, Cursor, Glean, Claude-3.7, Claude-3.5, Figma.

What results were reported?

Time savings on frontend workflows: up to 20%; feature flags cleaned up in a single PR: 15+; Fizz time to market: concept to customer-ready in just a few short months; task duration for iOS search feature: tasks that traditionally took several days were completed in hours (source-reported, not independently verified).

What failed first in this deployment?

AI coding agents struggled with large legacy monolith codebases where unclear boundaries led to slower suggestions, incomplete answers, or hallucinated code.

How is this quality assurance AI workflow structured?

AI tools embedded in daily tasks → AI agent code cleanup → Figma-to-scaffold generation → AI-assisted debugging → RCA documentation generation → Prompt engineering playbooks.

Related quality assurance cases