Benchmark report

# The Global 1000 Cloud Agent Benchmark

Five months of production data from Global 1000 engineering teams running cloud agents. See what usage, cost per outcome, and output look like at scale before you set your own agent strategy.

- Cost per session fell 21% while session volume grew 147%
- Productive hours spread across every SDLC stage, not just Build
- Build and Test take 66% of spend and produce 74% of merged PRs

![The Global 1000 Cloud Agent Benchmark cover](https://devin.ai/_next/image?url=%2F_next%2Fstatic%2Fimmutable%2Fmedia%2Fcover.12bq56d47w8d1.png\&w=3840\&q=75)

PDF · 7 pages · 10 min read

Get the report

## What’s inside

1. 01

   ### Cloud agents become more efficient with scale

   Sessions, monthly active users, and dollar cost per session and per merged PR, indexed to the first month of the period.

2. 02

   ### Why efficiency improves as cloud agents scale

   Centralized enablement, governance, and compounding playbooks versus local tools and usage caps.

3. 03

   ### Cloud agents accelerate the entire SDLC

   Share of sessions and productive hours across Plan and Design, Build, Test, Secure, and Deploy.

4. 04

   ### Spend and output focus on high-value work

   Where ACUs and merged PRs concentrate, and productive hours per session ranked by use case.

## What you’ll learn

- How cost per outcome moves as agent usage grows past the pilot stage
- Which SDLC stages generate productive hours, and how evenly
- Where enterprises point agent spend, and what it returns in merged code
- Which use cases return the most engineering hours per session
- How a session, a productive hour, and an ACU are measured

[Get the report →](#cloud-agent-benchmark-form)

> “Deployed well, cloud agents show real economies of scale: cost per output falls while usage and the value delivered keep growing.”
>
> Cognition Executive Briefing

## Get the full benchmark

The same metrics our product and applied AI teams use with customers, with methodology notes for each figure.