Quality assurance · Production

What Cognition Learned Building Cloud Agents for Enterprise Engineering Orgs

The problem

Building cloud agent infrastructure requires solving three interconnected challenges that containerized approaches cannot address: shared-kernel security threats, inability to persist agent state across the async gaps of real engineering work, and the massive orchestration and governance investment needed to operate at enterprise scale.

First attempt

A leading cloud data platform company attempted to build in-house cloud agent infrastructure but abandoned the project because the combined scope of orchestration, governance, and integrations overwhelmed their infrastructure team.

Workflow diagram · grounded in source
1
Engineer delegates task
trigger
“Engineers need to learn which work to delegate and which to keep, and how to define tasks precisely enough that agents execute without constant correction.”
2
Isolated VM execution
ai_action
“every agent session runs on its own dedicated kernel with fully isolated storage, networking, and compute”
3
CI and source control integration
integration
“An agent opens a PR, waits on CI, responds to code review, reruns tests, and pushes a follow-up commit.”
4
State snapshot and async resume
feedback_loop
“Compute shuts down while the agent is idle, and the session resumes exactly where it left off when a CI result or review comment arrives.”
5
Engineer review
human_review
“The volume of code that needs review increases dramatically, but the review process designed for human-authored code doesn't transfer cleanly. Teams need to establish what rigorous review looks like for agent-produced work at a much high…”
Reported outcome

Itaú, the largest private bank in Latin America, deployed cloud agents across nearly 17,000 engineers and after eleven months completed migrations 5 to 6x faster, auto-remediated 70% of static-analysis security vulnerabilities, and increased test coverage by 2x.

Reported metrics
Migration speed improvement5 to 6x faster
Static-analysis security vulnerabilities auto-remediated70%
Test coverage increase2x
Engineers in deploymentnearly 17,000 engineers
Reported stack
DevinmicroVMs
Source
https://cognition.ai/blog/what-we-learned-building-cloud-agents
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Itaú, the largest private bank in Latin America, deployed cloud agents across nearly 17,000 engineers and after eleven months completed migrations 5 to 6x faster, auto-remediated 70% of static-analysis security vulner…

What tools did this team use?

Devin, microVMs.

What results were reported?

Migration speed improvement: 5 to 6x faster; Static-analysis security vulnerabilities auto-remediated: 70%; Test coverage increase: 2x; Engineers in deployment: nearly 17,000 engineers (source-reported, not independently verified).

What failed first in this deployment?

A leading cloud data platform company attempted to build in-house cloud agent infrastructure but abandoned the project because the combined scope of orchestration, governance, and integrations overwhelmed their infras…

How is this quality assurance AI workflow structured?

Engineer delegates task → Isolated VM execution → CI and source control integration → State snapshot and async resume → Engineer review.

Related quality assurance cases