How Delta Air Lines scaled AI-native development to 1,200 engineers with cloud development environments

When Delta Air Lines' Developer Experience team started measuring its Net Promoter Score (NPS), they faced a brutal reality: It was very low. Two years later, after deploying cloud development environments (CDEs) to more than 1,200 developers, that same team achieved:

• NPS improvement.
• Performance gains for developers switching from virtual desktop infrastructure (VDI). 
• Reduced onboarding time from hours and weeks to minutes.

Delta built its platform with Red Hat OpenShift Dev Spaces (cloud development environments) to solve traditional development problems like onboarding, dependency conflicts, and governance. Then AI-native development emerged, and that same platform became critical in a way the company hadn't anticipated.

At Red Hat Summit 2026, Vidheer Gadikota, General Manager of IT at Delta Air Lines, joined Red Hat Architect Charro Lee Gruver and Senior Director of Developer Tools Balaji Sivasubramanian to share how Delta's infrastructure enabled AI agents at enterprise scale. This is the story of how they built it—and why the platform they created for conventional challenges was perfectly positioned for AI-native development.

Why Delta started with cloud development environments

Before AI agents entered the picture, developers needed to address dependency conflicts on local workstations, a process that could take days. In turn, this set-up work took away from engineers being able to innovate and provide value to Delta more broadly. 

Imagine a developer working on a new application built with the latest Java version. They are assigned a maintenance ticket for a legacy application running on Java 8—a version from 2014 that's no longer supported. They then spend hours reconfiguring their laptop to switch between incompatible environments instead of actually fixing the bug.

Running development environments locally created several problems for Delta: 

• Onboarding new developers with project environments often took weeks because of the time required to properly configure them.
• Local development environments made governance of source code and credentials harder.
• Windows machines couldn't run the Linux-based, containerized workflows that cloud-native development requires.

Delta piloted cloud development environments with 20 engineers to test, learn, and prove it could work at scale, and launched with successful week-long, hands-on experiences. Developers finally had autonomy without compromising security.

Then AI-native development arrived.

Laptops become a bottleneck and more vulnerable with AI

A couple of months after Delta launched an AI-native integrated development environment (IDE), the team discovered a senior developer was using AI to solve 7 days’ worth of complex problems by lunchtime. But there was a dilemma: He couldn’t use the laptop for other tasks, like sending an email, while the AI agent was working. The agent used 100% of the CPU and memory.

This wasn't an isolated event. As Delta's engineers adopted AI-native tools, 3 critical challenges emerged.

Compute demands exploded

AI agents work continuously, not just when you're actively coding. They run build processes, test suites, and environment simulations in the background. Developers reported leaving laptops open overnight with chargers plugged in, unable to close them or switch networks without killing agent progress. The laptop became a constraint, not an enabler.

Security surfaces expanded

AI-native IDEs add new integration points through AI inference endpoints and large language model (LLM) integrations. New extensions appear constantly, and validating them takes time. Local development made that work harder to govern.

Consistency became impossible

The same dependency conflicts that plagued traditional development were amplified with AI tooling. Teams couldn't reproduce agent behaviors when every developer had different setups, application programming interface (API) keys, model configurations, and Model Context Protocol (MCP) servers. Troubleshooting became guesswork.

Delta needed AI agents at scale, with enterprise security by default, running on elastic compute that didn't drain laptops.

And they already had the platform to solve it.

The solution: Red Hat OpenShift Dev Spaces for AI agents

The architecture Delta built for traditional development challenges solved AI-native development problems by design.

Developers use AI-native tools like Kiro, AWS's AI-native IDE, for the interface. The workspace—where code lives, builds run, and AI agents work—runs entirely in Red Hat OpenShift clusters using OpenShift Dev Spaces.

Workspaces as code 

The environment for each project—tools, libraries, frameworks, AI configurations, API keys, MCP servers—lives in the Git repository as a YAML file. This isn't defined once per developer; it's defined once per code base. When a developer starts work on any project, they get a workspace purpose-built for that code base in under 2 minutes. A Java 25 project has its workspace. A Java 8 legacy application has a different one. No more dependency conflicts. Each workspace is isolated, configured, and ready.

AI inference and agents in the cloud

Delta controls where models are hosted: self-hosted for air-gapped use cases, or approved cloud providers when that makes sense. API keys are managed centrally, not scattered across laptops.

Enterprise security by default

"The best security for any developer is the security they never have to work around," Gadikota said. 

Code stays within enterprise boundaries. Extensions are scanned and served from Delta's private registry. Secrets never touch local hard drives.

This delivers what AI-native development requires:

• Always-on agents: Developers can close laptops, switch networks, and go to lunch while agents keep running in the cloud. It's a Kubernetes workload, not a local process.
• Elastic resources on demand: Graphics processing unit (GPU) nodes and additional compute are available without filing tickets. No laptop thermal throttling.
• Multiple agents, multiple projects: Developers spin up separate workspaces and run agents in parallel, not limited by laptop resources.

Enabling AI agents safely and at scale

As Delta scaled cloud development environments for AI-native workflows, they hit a technical barrier that most enterprises haven't encountered yet.

AI agents don't just write code. Behind the scenes, agents and subagents need to run build processes, execute unit tests, and simulate environments to prove solutions work. To do all of this autonomously, agents need to spin up containers—inside the workspace container that's already running in the cloud.

Nested containers. Containers within containers.

"Which means the users need to have privileges and root access to do that," Gadikota explained to the audience at Red Hat Summit. "If anyone from InfoSec is in this room, you must have flinched a bit because I said that."

Root access to the entire cluster just to enable AI workflows? That's a security nonstarter. But without it, AI agents can't do their job.

This is where Delta's collaboration with Red Hat became critical. It also illustrates what working with an open source vendor looks like in practice.

Gruver, the Red Hat architect embedded with Delta's team, started building a solution to enable permissions at the namespace level rather than root access for the whole cluster. This was created before the agentic AI moment to let development tools run containers within a developer’s isolated workspace. But it also enabled AI agents to do the same. Delta's architect, Adnan, built the integration to connect their AI IDE with OpenShift Dev Spaces. They experimented with it, gave feedback to Red Hat Engineering, and Red Hat made the capability available in Red Hat OpenShift 4.17 onwards.

"We are contributing back to Red Hat," Gadikota noted, "so everyone will benefit from it."

Now developers can run Podman, Docker, test containers, Ansible deployments, and Tekton functions all within their AI-enabled workspace, with AI agents spinning up whatever they need to validate solutions. No root access required. No security exceptions filed.

"If you partner with Red Hat, this is not like a vendor-customer relationship," Gruver explained. 

The solution Delta needed for AI-native development at scale became a capability the entire OpenShift community can use.

Platform for AI-native development, not just a tool

Delta's most strategic insight: Invest in the platform, not just the tool.

"This is very important," Gadikota emphasized, "because the products or the tools that are offered today will no longer be available down the line in a month or in a year. The product that we are using today is going to take 3 more different flavors, and a completely new offering will come up."

AI tooling evolves fast. New AI assistants appear weekly. The IDE Delta uses today will look different in 6 months. Model capabilities advance rapidly.

"But what's most important is having that platform mindset,” continued Gadikota. 

This is why Delta built on OpenShift Dev Spaces, a native OpenShift workload integrated with their existing GitHub, GitLab, continuous integration and continuous delivery (CI/CD) pipelines, and security tooling. When new AI-native IDEs emerge, Delta can integrate them without rearchitecting their entire developer experience. The workspace definitions, security posture, governance model, and compute infrastructure all remain consistent. Only the user interface (UI) layer changes.

4 takeaways for allowing AI-native development

1. Meet developers where they are. Enable the AI-native IDEs they want within your governance model. Cloud development environments give developers autonomy while maintaining enterprise security.

2. Make security the default, not the obstacle. Run AI agents in governed environments where security is built in, not bolted on. The best security is invisible to developers.

3. Gamify adoption, don't mandate it. Product launches with hands-on experiences work. Email mandates die in inboxes.

4. Invest in the platform, not just tools. Build a platform that can integrate whatever AI tooling emerges next, so your adoption continues regardless of which specific assistant wins the market.

Delta's transformation started by solving traditional development problems with CDEs. When AI-native development emerged with new challenges—agents using 100% of laptop compute, new security surfacing, consistency nightmares—their existing platform was ready.

For Delta's thousands of engineers operating more than 5,000 daily flights and serving 200 million passengers, the shift from fighting laptop constraints to delivering customer value matters.

The future of development is AI native. The infrastructure to support it needs to be cloud native.

Ready to deliver applications faster by improving your developer experience?

• Try Red Hat OpenShift Dev Spaces: Discover how to give your developers zero-setup workspaces with centralized security controls. Start a self-service trial.
• Provide an agent-ready developer portal: Giving developers and AI agents organizational context—services, documentation, and golden paths—lets them build, deploy, and operate applications within enterprise standards. Get started with Red Hat Developer Hub.
• Uncover what's slowing developers and blocking AI adoption: Identify hidden inefficiencies—from onboarding friction to toolchain sprawl—and build a roadmap that alleviates traditional developer obstacles and facilitates AI-native development at enterprise scale. Learn more about the assessment.