Enterprise report

# Smarter AI Caps for Engineering

Token limits set by efficiency, not blanket rules. See how Efficiency Score, Devin Coach, and efficiency-based approvals let enterprises allocate AI spend by engineering efficiency instead of uniform usage caps.

- How a 0–100 Efficiency Score grades each engineer on productive hours per unit of spend
- Why one engineer's cap was lifted 22× automatically while merged PRs rose from 30 to \~80 a week
- How Devin Coach flags vague prompts, expensive modes, and mixed-up sessions before they run

![Smarter AI Caps for Engineering cover](https://devin.ai/_next/image?url=%2F_next%2Fstatic%2Fimmutable%2Fmedia%2Freport.442etocs4j_1_.png\&w=3840\&q=75)

PDF · 7 pages · 10 min read

Get the report

## What’s inside

1. 01

   ### Giving each engineer an Efficiency Score

   How Devin credits work that never becomes a PR, estimates productive engineering hours, and combines efficiency and volume ratings into a 0–100 score, with four worked examples.

2. 02

   ### Guiding engineers before they spend with Devin Coach

   The prompt-level suggestions Devin gives before a session runs: unrelated follow-ups, parallelizable tasks, expensive modes for simple work, and non-work usage.

3. 03

   ### Raising limits automatically with efficiency-based approvals

   How admins set an auto-approval ceiling, which grades get automatic increases, and what happened when two top engineers at different companies hit their monthly cap.

4. 04

   ### Allocating spend where it has the most impact

   Why blanket caps act like a hiring freeze in the AI era, and what comes next: budgets delegated to department leaders and sub-teams.

## What you’ll learn

- How to measure AI output in productive engineering hours instead of merged PRs alone
- How the Efficiency Score weights efficiency over volume, and why a careful low-volume engineer can outscore a power user
- What admins see when they review a usage request: efficiency, model mix, and consumption
- How to lift caps for your best engineers without adding to admins' review queue

[Get the report →](#ai-efficiency-report-form)

> “Devin's intelligent cost controls have helped us keep a lid on runaway agent spend. Efficiency-based auto-approval has really eased the burden on our admins. Our power users get more usage without having to ask, and when someone's usage isn't paying off, we find out and can have a conversation with them.”
>
> Will Laughlin, VP of Engineering, Bilt

## Allocate AI spend where it has the most impact

Download the report to see how enterprises replace blanket usage caps with efficiency-based limits, coaching, and automatic approvals.