# Acceleration without chaos

Devin gets end-to-end engineering work done across your SDLC,\
with the trust and ROI an enterprise requires

[Contact Us](#contact)[Review security](#security)

-

* ## [Model Wisdom](#model-wisdom)
  [Every frontier model, plus an opinion on which is right for the task at hand ](#model-wisdom)
* ## [Outcomes Win](#earns-its-keep)
  [Tasks over tokens. Value over spend. Have visibility of output and ROI ](#earns-its-keep)
* ## [No Shadows](#trust)
  [The access, budget, and security controls that you’ve been missing ](#trust)
* ## [Level Up](#level-up)
  [Your team will set a new pace for the company by working with our FDEs ](#level-up)

![Citi](https://devin.ai/images/customer-logos/citi.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

[](https://devin.ai/customers/rvtech)

[](https://devin.ai/customers/itau)

[](https://cognition.com/blog/mercedes-benz-cognition)

![Lowe's](https://devin.ai/images/customer-logos/lowes.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

![Palantir](https://devin.ai/images/customer-logos/palantir.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

[](https://devin.ai/blog/devin-for-mongodb-modernizations)

![Cisco](https://devin.ai/images/customer-logos/cisco.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

[](https://www.goldmansachs.com/insights/talks-at-gs/russell-kaplan)

![ServiceNow](https://devin.ai/images/customer-logos/servicenow.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

![Hertz](https://devin.ai/images/customer-logos/hertz.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

![Cloudflare](https://devin.ai/images/customer-logos/cloudflare.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

![Elevance Health](https://devin.ai/images/customer-logos/elevance.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

![Fox](https://devin.ai/images/customer-logos/fox.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

## Citi on Devin

Jane FraserCEO, Citi![Jane Fraser](https://devin.ai/_next/image?url=%2Fimages%2Fenterprise%2Fjane-fraser.webp\&w=3840\&q=75\&dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

> “14,000 of our engineers use our main agentic coding tool, Devin, making it much more efficient to do work in-house. We've been in-sourcing work and have reduced our tech vendor spend by 30% year-over-year as a consequence, with a long way still to run”

[Citi Q3 2026 earnings call](https://www.citigroup.com/rcs/citigpa/storage/public/Earnings/Q32026/3q26-citigroup-earnings-transcript.pdf)![Citi](https://devin.ai/images/enterprise/citi-logo.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

Model Wisdom

01

## Let the models compete. Devin gets the best from all of them

Every frontier model from Anthropic, OpenAI and Google, plus Cognition's own SWE-2, under one contract. Whichever lab pulls ahead next, your engineers already have it

- ### One contract

  One agreement, one bill, one security review, every model

- ### Smart routing

  Each task goes to the model that fits it. Nobody has to keep up with every lab

- ### Fusion

  A frontier model leads, an efficient one does the work. Your engineers see one Devin

![](https://devin.ai/_next/image?url=%2Fassets%2Fimages%2Fnew-home%2Fgraphic-backgrounds%2Fbg-6.jpg\&w=3840\&q=75\&dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

Outcomes Win

02

## Tasks over tokens. Value over spend

Every finished task is scored in engineering hours, so value against spend is a number on a dashboard, not a feeling. And the price tracks the work, not the model behind it

- ### One unit

  An agent compute unit (ACU) counts everything a task used, normalized across models, so the price holds as models change

- ### More tasks per ACU

  The same output needed up to 40% fewer ACUs in September, from new models, better routing and a leaner harness

- ### Spend on what ships

  Devin Coach flags wasteful prompts before they are sent, so budget goes to work that ships

![Fable 5.1 as lead sends a brief to SWE-2 as sidekick, which returns results and gets feedback](https://devin.ai/_next/image?url=%2Fimages%2Fenterprise%2Flead-sidekick.webp\&w=3840\&q=75\&dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

Up to 60%

lower cost for frontier-model results with Devin FusionFrontierCode Extended, August 2026

## Up to $10M in guaranteed engineering value

If Devin delivers less engineering value than you spent over the year, we cover the gap in usage credits. Measured in engineering hours on your own work; a session whose pull request never merged counts for nothing

[How the AI Productivity Guarantee works](https://devin.ai/guarantee)

No Shadows

03

## Nothing unapproved, nothing unseen

Shadow AI is the tool nobody approved and the spend nobody can see. Devin sets access, budgets and integrations once, holds them on every surface, and logs every action

- ### Access

  SAML and OIDC single sign-on, SCIM provisioning, and role-based access with fine-grained permissions and custom roles

- ### Models and commands

  Allowlists for models, MCP servers and commands on Desktop and CLI: an enterprise baseline, overrides per organization

- ### Integrations

  Source control, issue trackers and MCP connections shared centrally, set up once, used by every organization you allow

- ### Audit

  Every session, setting and action in the audit log, with exports and an API

![](https://devin.ai/_next/image?url=%2Fassets%2Fimages%2Fnew-home%2Fgraphic-backgrounds%2Fbg-4.jpg\&w=3840\&q=75\&dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

Level Up

04

## Forward Deployed Engineers from the greatest companies in the world

Included in every enterprise contract, no upsell, no services firm in between. Engineers who work inside your environment on the outcomes you are measured on, and deliver every quarter

- ### White-glove deployment

  Devin configured and integrated with your security and governance, alongside your platform team, before the first project starts

- ### Your most important outcomes

  Key projects identified together and co-owned by your FDE, with the return measured against a baseline

- ### Best practices, company-wide

  Onsite enablement and workshops that train the next team on what worked for the first

- ### Top FDE talent

  Experience from Palantir, Tesla, Google, Waymo, Scale and Modal, applied to your team's needs

![A forward deployed engineer pointing at code on a monitor beside an engineer](https://devin.ai/_next/image?url=%2Fimages%2Fenterprise%2Fforward-deployed.webp\&w=3840\&q=75\&dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

- fewer ACUs per PR after enablement

  29%

- more sessions end in a merged PR

  39%

AI Analytics

05

## Know what you got, and what it cost

Each task carries its engineering-hours score. See value against spend by team, repository and project, on a record that cannot be edited after the fact

- ### Engineering hours

  Output measured as the time the work would have taken by hand, by an estimator built to undercount

- ### Session efficiency

  Sessions with an outcome, ACUs per merged pull request, ACUs per session, and how efficiency has changed over time

- ### What the work was

  Every session typed by kind of work, feature to bug fix, with the hours each returned by organization and surface

![](https://devin.ai/_next/image?url=%2Fassets%2Fimages%2Fnew-home%2Fgraphic-backgrounds%2Fbg-4.jpg\&w=3840\&q=75\&dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

AI Cost Controls

06

## Budgets that hold. Per organization, per tier, per person

Monthly allowances follow your identity groups: a new hire might start on 100 ACUs a month, an engineer on 1,000, a power user on more, and the limit holds until an admin says otherwise

- ### Usage tiers

  Monthly ACU allocations, ranked, with a default tier for everyone else and a budget per organization on top

- ### More when they need it

  Members ask for more inside Devin and admins approve or deny it there, or the request routes to your own intake tool

- ### Group mappings

  Identity-provider groups map to tiers, or define your own. In several groups, you get the highest-ranked tier

![](https://devin.ai/_next/image?url=%2Fassets%2Fimages%2Fnew-home%2Fgraphic-backgrounds%2Fbg-5.jpg\&w=3840\&q=75\&dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)

Security & Trust

07

## Built to pass your security review

Your code never trains a model and never leaves the boundary you set: isolated sessions, connected to your network, single-tenant and your region if you need them[Visit the Trust Center](https://trust.cognition.ai/)

- ### Your data stays yours

  No training on your code. Zero retention at model providers. Inputs and outputs are your IP, with attribution filtering

- ### Every session isolated

  Every session on its own machine with kernel-level isolation. Secrets stay out of the prompt. IP access lists for app and API

- ### Your network, your terms

  AWS PrivateLink or IPSec into your systems, plus Devin Connect, a policy gateway you run, reached by an outbound-only tunnel

* ![SOC 2 Type II](https://devin.ai/assets/images/new-home/soc-2.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)Certified on all surfaces
* ![ISO 27001](https://devin.ai/assets/images/new-home/iso-27001.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)Certified on all surfaces
* ![FedRAMP High](https://devin.ai/assets/images/new-home/fedramp.svg?dpl=dpl_5ZzaxXXChWhEcm5NXaaGWCvehM2U)CLI and Desktop Cloud in progress

- 6 of 8

  US global systemically important banks

- 3 of 6

  top US health insurers

- 3

  major US wireless carriers

## Frequently Asked Questions

### Our engineers already use Cursor and Claude Code. Why add Devin?

Keep them. A local tool makes one engineer faster. Devin adds capacity: it takes whole tasks end to end on its own cloud machines, runs them in parallel across repositories, and keeps going after everyone has logged off. Outcomes multiply, not just speed

### Do you train on our code?

No. Cognition never trains models on customer code, and zero data retention is enforced with model providers

### Can Devin reach systems inside our network?

Yes. Your environment connects to your network over AWS PrivateLink or IPSec, and Devin Connect reaches internal systems through a gateway you run, over an outbound-only tunnel. If your requirements are stricter, talk to us; we have met them before

### Which models can we use, and who decides?

Every frontier model is under your contract, and Devin routes each task to the one that fits it. Admins can restrict models on Desktop and CLI, restrict modes on Cloud, and try new models in preview first

### What is an ACU?

An agent compute unit: the normalized measure of what a task used, model and machine included, across every model. Budgets, caps and the guarantee all count in ACUs

### How does the Productivity Guarantee work?

Each finished task is scored in engineering hours. A month before your annual renewal, that value is compared with what you spent. If it falls short, we cover the gap in usage credits, up to $10M

### Does it work with self-hosted source control?

Yes: GitHub Enterprise Server, GitLab self-managed, Azure DevOps Server, Bitbucket Data Center, Gerrit and Perforce

### How do we buy?

Through a direct agreement, or through AWS, Microsoft Azure, Google Cloud, Databricks and MongoDB, so the purchase draws down your committed spend

### What does a pilot look like?

A few weeks on a handful of use cases tied to your biggest pain points, with success criteria agreed up front and access settled before kickoff. Our engineers work alongside yours, and the readout compares the cost in ACUs with what the work costs you today

## See how Devin can work for your team

Tell us about your team and your codebase. We'll show you what Devin can take on, under your rules