Most companies say they’re “embracing AI.” Endava is doing something more specific — and more interesting. The London-based IT services firm, which employs around 11,000 people and works with clients across financial services, healthcare, and retail, is restructuring how it actually delivers software around AI agents, ChatGPT Enterprise, and OpenAI Codex. Not as a side experiment. As the core operating model. OpenAI published a detailed case study on June 4, 2026, walking through what that looks like in practice — and it’s one of the more concrete enterprise AI stories I’ve seen in a while.
Why Endava, and Why Now?
Endava sits in an interesting position. It’s not a tech giant with unlimited R&D budget, and it’s not a scrappy startup that can pivot overnight. It’s a mid-size professional services firm where the product is, essentially, human expertise applied to software problems. That’s exactly the kind of business where AI creates the most tension — and potentially the most opportunity.
The company has been dealing with the same pressures everyone in IT services faces: clients want faster delivery, tighter budgets, and more predictability. At the same time, the talent market for skilled engineers hasn’t gotten easier. So the question Endava’s leadership seems to have asked isn’t “how do we use AI to cut costs” but rather “how do we use AI to deliver outcomes we couldn’t deliver before?”
That framing matters. A lot of enterprise AI deployments fail or stall because they’re cost-reduction stories dressed up as innovation. Endava’s approach, at least from what OpenAI has shared, is built around capacity expansion — doing more, faster, with the same or growing teams. That’s a more sustainable pitch internally, and it shows in how they’ve structured the rollout.
What They Actually Built
The Endava deployment isn’t a single tool. It’s a stack of interconnected capabilities built on top of OpenAI’s enterprise offerings, deployed across different parts of the software delivery lifecycle.
ChatGPT Enterprise as the Knowledge Layer
ChatGPT Enterprise is being used broadly across the organization — not just by engineers, but by project managers, delivery leads, and business analysts. The use cases here are fairly practical: drafting client communications, summarizing long technical documents, generating status reports, and helping non-technical stakeholders get up to speed on complex codebases faster. It’s the connective tissue between different roles on a project team.
What makes this more interesting than a standard ChatGPT rollout is the emphasis on internal knowledge integration. Endava has been working to connect ChatGPT Enterprise to its own proprietary documentation, past project data, and delivery frameworks — so the model isn’t just drawing on general knowledge but on Endava-specific context. That’s where the real productivity lift tends to come from.
Codex for Agentic Code Work
OpenAI Codex is getting used more aggressively. Rather than just autocomplete or inline suggestions, Endava is deploying Codex in agentic configurations — where the model can take a task, execute a series of steps, write and test code, and return a result without a human approving every micro-decision along the way.
This is where the “redesigning software delivery” framing becomes real. In traditional delivery, a developer gets a ticket, writes code, submits a PR, waits for review. With agentic Codex workflows, a chunk of that cycle — particularly the boilerplate-heavy, well-defined parts — can run largely autonomously. Endava is using this for things like scaffolding new services, writing unit tests, performing code migrations, and generating documentation from existing code.
- Automated test generation: Codex agents writing test suites based on existing code, reducing manual QA effort significantly
- Code migration assistance: Moving legacy codebases to newer frameworks with agent-assisted refactoring
- Documentation generation: Auto-generating technical docs directly from source code, keeping them in sync with the actual implementation
- Sprint scaffolding: Generating boilerplate code for new features based on existing architectural patterns in a project
- Workflow automation: Agents handling repetitive integration tasks that previously required manual developer time
Building an AI-Native Culture
The piece I find most telling in OpenAI’s case study is the emphasis on culture change. Endava isn’t just dropping tools into existing workflows — they’re training people to think differently about what their job is. Engineers are being encouraged to act more as orchestrators and reviewers of AI output rather than primary producers of every line of code. That’s a meaningful shift, and it doesn’t happen automatically.
Endava has apparently invested heavily in internal enablement: training programs, prompt engineering guidelines, and what they’re calling an “AI-native” delivery framework. The goal is that AI agent usage becomes default behavior, not an advanced option for the most tech-curious employees. That’s the right instinct — the companies that get the most out of these tools are the ones that make them unavoidable, not optional.
What This Actually Means for Enterprise Software Delivery
Here’s the thing: Endava is an IT services company. Its clients are paying for outcomes — working software, delivered on time, within budget. So when Endava integrates AI agents into its delivery pipeline, the downstream effect is that those clients get faster timelines and potentially lower costs without having to manage any of the AI complexity themselves. The AI abstraction happens inside Endava’s operation.
That’s a genuinely interesting business model shift. Endava’s competitors — Accenture, Infosys, Wipro, Cognizant — are all making similar moves, but the speed and specificity with which Endava has integrated OpenAI’s stack is worth watching. If they can consistently demonstrate 20-30% faster delivery cycles (which agentic coding workflows have shown is achievable in controlled settings), that becomes a competitive differentiator in a margin-compressed industry.
There’s also a workforce question lurking here. Endava employs thousands of engineers. If agentic AI can absorb a meaningful percentage of coding work, what happens to headcount? The company’s current framing — capacity expansion rather than reduction — is the politically safe answer, and it may well be true in the near term. But I wouldn’t expect that framing to hold indefinitely as the tools get more capable. We’ve seen similar dynamics play out with automation in other industries, and IT services won’t be immune.
It’s also worth noting how this fits into OpenAI’s broader enterprise strategy. Endava is far from the only large organization building on ChatGPT Enterprise and Codex. Travelers deployed an AI claims assistant nationwide earlier this year, and MUFG is restructuring significant parts of its operation around OpenAI’s models. The pattern is consistent: large organizations with complex, documentation-heavy workflows are finding that ChatGPT Enterprise plus task-specific agents creates compounding returns across departments.
The competition here is real. Codex is now available on AWS, which opens up enterprise deployments that need to stay within Amazon’s infrastructure. GitHub Copilot Workspace is making its own push into agentic coding. Google’s Gemini Code Assist is showing up in more enterprise pitches. The race to own the agentic software development workflow is genuinely competitive, and today’s case studies are tomorrow’s procurement decisions.
What This Means for Different Audiences
If you’re an enterprise IT leader, Endava’s model is worth studying as a template. The key insight isn’t the specific tools — it’s the delivery framework redesign. You can’t just add AI tools to an existing process and expect compounding returns. You have to restructure the process around what AI is actually good at.
If you’re a software engineer at a services firm, this is a preview of what your job looks like in two to three years. The engineers getting promoted won’t necessarily be the fastest coders — they’ll be the ones who are best at directing, reviewing, and debugging agent output. That’s a different skill set, and it’s worth developing now. OpenAI has been pushing Codex toward broader accessibility, which means the bar for working with these tools is dropping fast.
If you’re a client of an IT services firm, the honest question to ask your vendor is: where exactly in my project are AI agents involved, and what’s your quality control process? The efficiency gains are real, but so is the risk of agent-generated code that looks right and isn’t. The firms that are transparent about this — like Endava appears to be — are the ones worth working with.
Frequently Asked Questions
What is Endava using OpenAI’s tools for?
Endava is using ChatGPT Enterprise for cross-functional knowledge work — drafting, summarizing, and internal communication — and OpenAI Codex in agentic configurations for automated coding tasks like test generation, documentation, and code migration. Together, they form a restructured software delivery pipeline rather than a collection of isolated tools.
How does agentic coding actually work in practice?
Rather than a developer writing every line of code manually, an AI agent receives a defined task, plans a series of steps, writes and executes code, runs tests, and returns a completed result. A human engineer reviews and approves the output rather than producing it from scratch. Codex has evolved significantly in its ability to handle multi-step agentic tasks with minimal supervision.
Is this approach unique to Endava or part of a broader trend?
It’s a broader trend, but Endava is among the more advanced documented examples specifically in IT services. Financial services, healthcare, and insurance companies are making similar moves with OpenAI’s enterprise stack, but the application to software delivery pipelines — where the output is code itself — is particularly significant for this sector.
What are the risks of this model?
The main risks are code quality and over-reliance on agents for tasks that require deeper contextual judgment. Agent-generated code can pass unit tests and still contain logic errors or security vulnerabilities. The firms getting this right are investing as much in human review processes as they are in the AI tooling itself — the two have to evolve together.
Endava’s bet is that the firms which figure out agentic delivery first will have a structural advantage that’s hard to close. Given how fast Codex and similar tools are improving, that window for differentiation may be shorter than they’d like — which means the pressure to move fast and move smart is only going to intensify across the industry.
