Endava and AI agents: how software delivery is changing

Endava: Chronology of a Silent Transformation

Endava is not a name that often comes up in conversations among Italian SMEs. However, it is one of the most significant digital engineering companies globally. Headquartered in London with operations spread across Europe, Latin America, and North America, the company manages complex projects for clients in the finance, healthcare, and retail sectors.

During 2025, Endava launched a structured AI agents adoption program. The stated objective was precise: to reduce software time-to-delivery without sacrificing quality. The program was developed in progressive phases, starting with high-volume, low-decision complexity use cases.

Therefore, the starting point was not the most advanced technology available. It was the identification of real bottlenecks in the development process. This methodological choice is, perhaps, the most useful lesson for those observing the case from the outside.

Tools at the core: ChatGPT Enterprise and Codex

The case study published by OpenAI Endava's technological architecture is centered around two core tools: ChatGPT Enterprise and Codex. This setup forms the backbone of their development and operational processes, enabling advanced capabilities in code generation, analysis, and client solutions. **Core Components:** * **ChatGPT Enterprise:** This forms the primary AI reasoning and generation engine. It's utilized for a wide range of tasks, including: * **Code Generation and Refinement:** Generating boilerplate code, writing unit tests, suggesting code optimizations, and even creating entire code snippets based on natural language descriptions. * **Documentation Assistance:** Automatically generating and updating technical documentation, API descriptions, and user guides. * **Natural Language Understanding (NLU) for Client Interaction:** Processing and understanding client requirements expressed in natural language, facilitating more intuitive and efficient communication. * **Design Pattern and Solution Exploration:** Assisting in brainstorming and exploring different architectural patterns or solutions to complex problems. * **Debugging and Error Analysis:** Analyzing error logs and providing potential explanations or solutions. * **Knowledge Management:** Acting as an intelligent knowledge base, recalling and synthesizing information from vast datasets for internal use and client-facing products. * **Content Generation:** Creating marketing copy, project proposals, and internal training materials. * **Codex:** While ChatGPT Enterprise focuses on broader AI capabilities, Codex is specifically honed for code understanding and generation. It acts as a powerful assistant for developers, enhancing productivity through: * **Advanced Code Completion:** Providing highly accurate and context-aware code suggestions that go beyond simple keyword completion. * **Code Translation:** Converting code from one programming language to another. * **Code Explanation:** Breaking down complex code segments into understandable natural language explanations. * **Bug Detection and Pattern Recognition:** Identifying potential bugs and common programming errors. * **API Integration Assistance:** Helping developers understand and integrate with various APIs more efficiently. * **Scripting and Automation:** Generating scripts for repetitive tasks and automation workflows. **Integration and Workflow:** The power of Endava's architecture lies in the seamless integration of ChatGPT Enterprise and Codex within their development lifecycle. This integration is facilitated through: 1. **Developer IDEs and Tools:** Both ChatGPT Enterprise and Codex are likely integrated directly into the Integrated Development Environments (IDEs) used by Endava's engineers. This allows for real-time assistance as developers write code. Plugins and extensions provide access to the AI capabilities directly within the coding environment. 2. **Internal Development Platforms:** Endava may leverage internal platforms or custom-built tools that orchestrate the use of these AI models. These platforms could manage API calls, handle data preprocessing, and aggregate results from both ChatGPT Enterprise and Codex to provide a unified experience. 3. **Client Solution Development:** For client-facing solutions, ChatGPT Enterprise and Codex are employed to accelerate the development of intelligent features. This could involve building AI-powered chatbots, personalized recommendation engines, intelligent automation tools, or advanced data analysis platforms for clients. 4. **DevOps and CI/CD Pipelines:** These AI tools can be integrated into Continuous Integration and Continuous Deployment (CI/CD) pipelines to automate code reviews, generate documentation for new releases, and even assist in drafting release notes. 5. **API Orchestration Layer:** A crucial element would be an API orchestration layer that manages requests and responses to both ChatGPT Enterprise and Codex. This layer would handle authentication, rate limiting, and potentially query optimization to ensure efficient and cost-effective usage of the AI services. 6. **Data Pipelines and Training:** While Endava uses the pre-trained models of ChatGPT Enterprise and Codex, they might also have mechanisms to feed anonymized or curated project data back into these systems (where appropriate and permissible) to fine-tune models for specific client domains or internal best practices, thus improving future performance. This requires robust data governance and privacy protocols. **Benefits of this Architecture:** * **Accelerated Development Cycles:** Significant reduction in time spent on repetitive coding tasks, documentation, and debugging. * **Enhanced Code Quality:** AI assistance in identifying bugs, suggesting optimizations, and enforcing coding standards leads to more robust and efficient code. * **Innovation and Prototyping:** Faster exploration of new ideas and quicker prototyping of complex solutions. * **Improved Developer Experience:** Developers can focus on higher-level problem-solving and creative aspects of software engineering, rather than getting bogged down in mundane tasks. * **Scalability of Solutions:** The AI capabilities enable Endava to build and scale intelligent solutions for a wider range of client needs. * **Knowledge Dissemination:** AI can help democratize access to expertise and best practices within the organization. In essence, Endava's technological architecture is built on leveraging cutting-edge AI tools like ChatGPT Enterprise and Codex to augment human expertise, streamline development processes, and deliver innovative, high-quality solutions to their clients more effectively. This approach positions them at the forefront of technology services, embracing AI as a fundamental enabler of their business.

ChatGPT Enterprise has been integrated into developers' daily workflows. In particular, it has found application in generating technical documentation, writing automated tests, and reviewing existing code. Codex, on the other hand, has been used to accelerate code production in specific languages, reducing the time spent on repetitive tasks.

Beyond this, Endava has developed specialized agents for vertical tasks. For example, one agent dedicated to migrating legacy codebases, and another for automatically generating technical specifications from business requirements. Therefore, this is not a generic use of AI; it is a targeted orchestration of specific capabilities.

According to the estimates reported in the case study, certain phases of the development cycle saw time reductions in the range of 30–40%. These figures are consistent with the observations made by McKinsey in its research on developer productivity with generative AI.

Who won and who had to adapt

In every transformation of this kind, there are winners and those who must redefine their role. In the case of Endava, the distinction is clear.

Senior developers benefited most from automation. In fact, freed from low-value tasks, they were able to focus on architecture, critical review, and design decisions. Their qualitative output increased measurably.

Conversely, junior profiles have experienced a more complex transition phase. Tasks traditionally assigned to entry-level developers—writing boilerplate code, basic documentation, manual testing—have become the domain of AI agents. Therefore, Endava has had to redesign onboarding and training paths for this segment of professionals.

Finally, project managers have seen the nature of their work change radically. Managing dependencies between tasks has become partially automated. As a result, their value has shifted towards AI process governance and exception management.

The cultural dimension: the data that numbers don't capture

We of SHM Studio We believe that Endava's most original contribution is not technological. It's organizational.

The company has invested significantly in building an AI-native culture. This has meant widespread training programs, internal communities of practice, and—an often overlooked aspect—a governance system that defines when agents can act autonomously and when human supervision is necessary.

Similarly to what has been observed in other enterprise contexts, the main risk was not resistance to change. It was superficial adoption: using AI tools without modifying underlying processes. Endava chose the longer, but more solid, path. Thus, it achieved results that hold over time.

This approach is supported by research from Harvard Business Review on building an AI-ready culture, which identifies governance and training as critical success factors for enterprise adoption.

SHM Studio's Reading: What it Means for Italian SMEs

The Endava case is enterprise by definition. However, the dynamics it describes are also relevant for smaller businesses. Italian SMEs operating in B2B or digital retail contexts are now faced with similar choices, on a different scale.

First, the logic of AI agents does not require large corporate infrastructure to be applied. Tools such as AI solutions already available on the market allow the automation of specific workflows — from content management to lead qualification — with accessible investments.

Secondly, the lesson about culture is directly transferable. A 50-person company that adopts AI tools without redefining roles and responsibilities will achieve marginal results. Conversely, a company that integrates AI into existing processes with methodology and training can gain significant competitive advantages.

Therefore, the starting point for an SME should not be the selection of the tool. It should be the mapping of high-volume, low-decision-complexity processes — exactly as Endava did.

Practical applications: from development workflows to marketing processes

The logic of AI agents applied by Endava to software development has a direct analog in digital marketing and communication processes. In these areas as well, there are high-volume, structured, and repetitive tasks that lend themselves to intelligent automation.

For example, SEO-oriented content production, digital campaign management, and periodic reporting are areas where AI agents can operate with reduced human oversight. Our services SEO e digital marketing already integrate these logics into operational workflows.

Additionally, content generation for LinkedIn campaign e Google Ads benefits directly from assisted automation. It's not about replacing human strategic judgment, but about accelerating the operational phases downstream of decisions.

In the same way, the copywriting The professional evolves towards a hybrid model: the professional defines strategy, tone, and objectives; the AI agent produces drafts and variations; the professional validates and refines. This model is already operational in many international agencies.

What can we expect in the next 18 months

The Endava case anticipates a direction that will become mainstream by 2027-2028. Some trends are already identifiable with sufficient clarity.

First of all, agent specialization will increase. Today's AI agents are relatively generalists. Later, we will see vertical agents for specific sectors—legal tech, healthcare, manufacturing—with reasoning capabilities adapted to their respective domains.

Furthermore, agent governance will become a critical competency. Companies that invest today in defining oversight policies and processes will have a structural advantage as system complexity increases.

Finally, the integration between AI agents and platforms of web development CMS will become standard. Companies that manually manage their digital presence today will be at a competitive disadvantage compared to those who have automated recurring operational processes.

To delve deeper into these topics or evaluate how to integrate AI logic into your organization's processes, it is possible Contact the SHM Studio team to explore related articles in our blog.