Endava Redesigns Software Delivery with AI Agents

Endava, a technology consulting firm with presence in more than 25 countries and clients in banking, insurance, and retail, published this week a detailed use case alongside OpenAI describing how it is reorganising its software delivery model around AI agents. This is not an isolated pilot: according to the article published on OpenAI's blog, the company has integrated ChatGPT Enterprise and Codex into workflows that previously relied almost entirely on human intervention at each step.

What stands out is not the use of AI tools—that is already common in the industry—but the decision architecture that Endava describes: agents that not only generate code, but coordinate phases of the software lifecycle, from effort estimation through pull request review and automated documentation.

What has actually changed

According to the use case, Endava has deployed specialized agents for concrete tasks within its engineering teams. Codex is used primarily to accelerate code generation and review, while ChatGPT Enterprise acts as a reasoning layer for requirements analysis and technical documentation synthesis tasks.

The model they describe is not that of a developer using a chat to resolve specific questions: it is a workflow where agents operate semi-autonomously over repositories and project management systems, and human engineers intervene at critical decision points. The company mentions reduced timelines in planning and testing phases, though it does not publish auditable figures in the article.

They also mention the cultural work: they have created what they call an "AI-native culture" within the organisation, with internal training, new roles for coordination between agents and teams, and processes to validate agent outputs before they reach production.

Why it matters for the industry

Endava is not a startup experimenting with emerging technology. It is a publicly traded company with over 11,000 employees whose core business is outsourcing and software development consulting. That a company in this position restructures its operating model around AI agents has direct consequences for how the industry understands staffing, hourly billing, and the value proposition of technology consultancies.

The question the article does not answer—and that the industry should ask itself—is what happens to the business model when agents absorb a significant portion of billable work. Endava presents AI as a capacity multiplier, but the economics of selling engineering hours clash with the logic of automating those same hours.

Who this matters for

This case is especially relevant to three profiles:

• CTOs and engineering directors at companies working with external consultants: it is becoming reasonable to ask what portion of delivered work is being generated by agents and how that is audited.
• Technology consulting professionals: the career model based on accumulating specialised hours is changing faster than internal training plans typically acknowledge.
• Product and architecture teams: the coordination between specialized agents and CI/CD workflows is a real design problem, not just a tooling matter.

OpenAI's article is, in part, marketing material—Endava is a client, and the "success story" format has its limitations as a source—but the technical description of the agent architecture is concrete enough to extract useful signals about where the technology services industry is headed.

From ClaudeWave, we read this as confirmation of something we have been seeing in integration projects for months: the shift from "code assistant" to "agent operating in the delivery cycle" is no longer theoretical, and consulting firms that do not have a clear position on this within the next twelve months will have to explain to their clients why they do not.