Most companies that adopt AI coding tools do it the same way: a few engineers get access, productivity ticks up a bit, leadership calls it a success. Endava, the London-headquartered IT services firm with over 11,000 engineers across 30-plus countries, decided to do something more ambitious. They didn’t just want faster code completion. They wanted to restructure how work actually flows through the organization — and they used OpenAI Codex to do it. The result, according to OpenAI’s published case study, is a company that has slashed requirements analysis time from multiple weeks down to just hours, and is actively rethinking what software delivery looks like when AI agents can carry real workloads.
Why Endava, and Why Now?
Endava sits in an interesting spot in the IT services world. It’s not as massive as Accenture or Infosys, but it punches above its weight in digital transformation work — particularly in financial services, healthcare, and consumer products. Its business model is fundamentally about delivering software faster and better than clients can do in-house. That model is under enormous pressure right now.
The rise of AI coding assistants has forced every IT services firm to ask a genuinely uncomfortable question: if AI can write code, what exactly are we selling? For Endava, the answer they’ve landed on is orchestration. Not just writing code, but structuring how agents, humans, and workflows interact across an entire delivery pipeline. That’s a meaningful strategic bet, and Codex is central to it.
The timing matters too. OpenAI launched Codex as a cloud-based software engineering agent in May 2025, positioning it as something meaningfully different from Copilot-style autocomplete. Codex runs tasks asynchronously in isolated cloud environments, can handle multi-step engineering work, and is designed to operate with significant autonomy. For a firm like Endava that manages large, complex client engagements, that’s a different kind of tool than anything that came before it.
What Endava Actually Built
The specifics of Endava’s Codex deployment are what make this case study worth examining closely. This isn’t a story about engineers using an AI assistant to write boilerplate faster. Endava has restructured parts of its delivery process around agentic workflows, with Codex handling tasks that previously required senior human time at the front end of projects.
Requirements Analysis in Hours, Not Weeks
One of the most striking claims in the case study is the compression of requirements analysis timelines. Traditionally, breaking down a client’s business requirements into actionable technical specifications is slow, labor-intensive work. Senior engineers and business analysts spend weeks interrogating stakeholders, resolving ambiguities, and producing documentation that development teams can actually use.
Endava reports that Codex agents can now handle significant portions of this process in hours. That doesn’t mean the humans are gone — it means the agents do the heavy lifting on initial analysis, surface the key ambiguities, and produce structured outputs that humans then review and refine. The cognitive load shifts. Engineers spend their time on judgment calls rather than information gathering.
Parallel Workstreams at Scale
Because Codex operates asynchronously, Endava can run multiple agent tasks simultaneously across different parts of a project. One agent might be analyzing a module’s test coverage while another drafts documentation and a third investigates a dependency conflict. This parallelization is where the real throughput gains come from — not any single task being faster, but many tasks running at once without blocking each other or requiring human coordination at each step.
Here’s the thing: this is exactly what makes agentic AI different from copilot-style tools. A coding assistant helps one engineer go faster. An agent framework helps an entire delivery organization do more work in parallel. Those are different problems with different solutions.
Key Capabilities Driving the Transformation
- Asynchronous task execution: Codex handles long-running engineering tasks in the background, freeing engineers to work on other problems rather than waiting.
- Isolated sandboxed environments: Each Codex task runs in its own cloud environment, which matters for enterprise clients with strict security requirements — code isn’t executing in a shared space.
- Multi-step reasoning on codebases: Codex can navigate large, unfamiliar codebases, understand context across files, and make changes that are coherent at the project level, not just the line level.
- Integration with existing workflows: Endava connected Codex to its existing tooling rather than building parallel systems, which reduced adoption friction considerably.
- Human-in-the-loop checkpoints: The agentic workflows aren’t fully autonomous — engineers review outputs at defined stages, which has been important for maintaining quality on client-facing work.
The Bigger Strategic Picture
What Endava is describing is a genuine organizational redesign, not just a tooling upgrade. That distinction is important, and I think it’s undersold in how most companies talk about AI adoption.
Most enterprise AI deployments follow a familiar pattern: identify a bottleneck, apply an AI tool to that bottleneck, measure the improvement, report success. Endava seems to be doing something harder — asking how the entire flow of work changes when agents can reliably handle categories of tasks that previously required experienced humans. That requires rethinking roles, handoffs, and quality controls, not just adding a new tool to the stack.
This also positions Endava directly against the narrative that AI will simply shrink IT services firms by automating their core product. Endava’s counter-argument, implicit in this deployment, is that firms that learn to orchestrate AI agents effectively will be able to take on more work, more complex work, and deliver it faster — making them more valuable to clients, not less relevant. Whether that argument holds long-term is genuinely unclear, but it’s a more sophisticated response than most IT services companies have managed so far.
It’s also worth watching how competitors respond. Cisco has made its own significant bets on Codex for enterprise engineering, and other large IT services firms are running their own pilots. The race isn’t really about which company adopts Codex first — it’s about which organizations actually restructure themselves around agentic workflows versus which ones just use AI to do the same work slightly faster.
How This Compares to Other Codex Deployments
Endava’s use case has some notable parallels to how other enterprises have approached Codex, but with meaningful differences. Ramp focused on cutting code review time — a more targeted, single-workflow application. Virgin Atlantic’s deployment centered on shipping a specific project on time. Endava’s ambition is broader: they’re trying to change the operating model of an entire services organization, which is a harder problem with potentially larger payoffs if it works.
The comparison to Virgin Atlantic’s Codex deployment is instructive. Virgin Atlantic had a concrete deadline and a specific codebase. Endava is managing dozens of simultaneous client engagements with different stacks, different requirements, and different risk tolerances. Scaling agentic workflows across that kind of heterogeneity is genuinely hard, and the fact that they’re reporting real results suggests they’ve solved at least some of the integration challenges that would stop most organizations cold.
What the Numbers Actually Tell Us
The headline metric — requirements analysis going from weeks to hours — is dramatic enough that it deserves some scrutiny. A few things to keep in mind: this likely applies to specific types of requirements work, not every client engagement equally. Complex regulatory or domain-specific requirements still need experienced human judgment. And “hours” probably means hours of agent runtime plus human review time, not just the clock ticking on a Codex task.
That said, even a partial compression of requirements timelines has significant commercial value for Endava. If they can start development faster on engagements, that’s a direct competitive advantage in pitching clients. And it frees senior engineers to spend time on the problems that actually require their expertise rather than the information-gathering work that precedes it.
What This Means for Enterprise AI Adoption
For enterprises watching from the sidelines, Endava’s deployment offers a few concrete lessons. First, agentic AI requires workflow redesign, not just tool deployment — if you’re just adding Codex to an unchanged process, you’re leaving most of the value on the table. Second, the parallelization benefits are real but require upfront investment in how tasks are structured and handed off between agents and humans. Third, the enterprises seeing the biggest gains aren’t treating AI as a cost-cutting measure; they’re treating it as a way to take on work they couldn’t previously handle.
- Start with high-volume, well-defined tasks where agent outputs are easy to verify — requirements analysis is a good fit precisely because humans can quickly spot when an output is wrong.
- Build review checkpoints into workflows rather than trying to run fully autonomous pipelines on client-facing work from day one.
- Measure throughput, not just speed — the real metric is how much more work the organization can take on, not just how fast individual tasks complete.
- Invest in prompt and workflow engineering as a genuine discipline, not an afterthought.
Endava’s story is one of the more honest enterprise AI case studies published this year — it’s specific about what changed, doesn’t oversell the autonomy of the agents, and is clear that humans remain central to the process. As more IT services firms face the same strategic pressure, expect to see a lot more organizations trying to replicate what Endava has built here. The firms that figure out how to do this at scale and across heterogeneous client environments will have a real advantage going into the next few years.
Frequently Asked Questions
What is Endava using OpenAI Codex for?
Endava is using Codex to build agentic workflows across its software delivery process, with a particular focus on requirements analysis and parallel task execution. The firm reports cutting requirements analysis time from weeks to hours by deploying Codex agents to handle initial analysis work that previously required significant senior engineer time.
How is Codex different from GitHub Copilot for enterprise use?
Copilot is primarily an inline coding assistant — it helps individual developers write code faster in their editor. Codex operates as an autonomous agent that can handle multi-step engineering tasks asynchronously in isolated cloud environments, making it suited for workflow-level automation rather than individual developer augmentation. For organizations like Endava managing large delivery pipelines, that distinction is significant.
Is Endava’s approach replicable for other IT services firms?
The core approach — structuring agent workflows around high-volume, verifiable tasks with human review checkpoints — is replicable, but it requires genuine investment in workflow redesign. Firms that try to bolt Codex onto unchanged processes won’t see the same results. The organizational change is as important as the technology choice.
What are the risks of running agentic workflows on client-facing engineering work?
The main risks are quality and accountability — if an agent produces incorrect analysis or code that a human reviewer misses, the error can propagate through the delivery pipeline. Endava’s approach of maintaining human-in-the-loop checkpoints at key stages is the standard mitigation, though it does limit how fully autonomous the workflows can be in practice.
