Endava Builds an Agentic Organization with Codex: Requirements Analysis Compressed from Weeks to Hours

Introduction: Leaping from Weeks to Hours

Global IT services company Endava is leveraging OpenAI's Codex to build an organizational architecture centered on Agentic AI, compressing requirements analysis in the software delivery process from weeks to hours. This transformation goes beyond a mere technological upgrade — it represents a cutting-edge practice in the evolution of enterprise organizational structures toward the "Agentic Organization."

What Is an Agentic Organization?

Organizational Transformation Beyond Tool Adoption

An "Agentic Organization" is not simply about embedding AI tools into existing workflows. It fundamentally redesigns how an organization operates. In this model, AI agents are no longer passive tools that respond to commands — they are "digital employees" capable of autonomously executing complex task chains. They can understand context, decompose tasks, make judgments, and deliver results.

This concept traces its roots to management theories of "Disintermediation" and "Business Process Reengineering" (BPR). When Michael Hammer introduced BPR in the 1990s, he emphasized using information technology as a lever to redesign business processes rather than merely automating existing ones. The Agentic Organization is a deepening of this philosophy in the AI era: AI agents don't just replace repetitive labor — they can take on knowledge work requiring judgment, exponentially expanding an organization's "cognitive bandwidth." From an organizational design perspective, Agentic Organizations exhibit flat, networked characteristics — AI agents are embedded as "digital nodes" in the organizational network, reducing information transmission losses inherent in traditional hierarchical structures. Endava's practice essentially achieves a dramatic reduction in coordination costs within the software delivery process, aligning closely with Nobel laureate Ronald Coase's transaction cost theory.

Endava's practice demonstrates that when AI agents are given sufficient autonomy with appropriate constraints, they can take on work that traditionally required collaboration across multiple teams — such as requirements analysis, code generation, and test case writing.

The Technical Core of Agentic AI

Agentic AI represents a critical evolutionary direction for large language models (LLMs). Unlike traditional single-turn Q&A-style AI, AI agents possess a closed-loop capability of "perceive-plan-execute": they can decompose complex goals into subtasks, invoke external tools (such as code executors, search engines, and database interfaces), and dynamically adjust strategies based on execution results. This capability relies on prompt engineering frameworks like ReAct (Reasoning + Acting) and Chain-of-Thought, as well as Function Calling and multi-step memory management mechanisms. It is precisely this autonomous "perceive-plan-execute" loop that enables AI agents to handle complex knowledge work that traditional AI tools cannot.

The Role of Codex in the Agentic Organization

OpenAI's Codex has gone through two distinctly different development phases. The early Codex (2021) was a code completion model fine-tuned on GPT-3, primarily serving as the underlying engine for GitHub Copilot with code auto-completion as its core function. The new-generation Codex released in 2025 is a complete AI software engineering agent platform built on the o3 model, capable of autonomously executing long-running, complex programming tasks in isolated cloud sandboxes. The new Codex supports running multiple independent tasks in parallel, can read and write files, execute commands, run test suites, and deeply integrate with existing development workflows through code hosting platforms like GitHub. This generational leap marks the evolution of AI programming tools from "assisted completion" to "autonomous delivery," providing the core infrastructure for Endava's Agentic Organization.

Notably, Codex's operation within cloud sandbox environments is a critical security design for enterprise deployment. Sandbox technology uses containerization (e.g., Docker) or virtual machine isolation to ensure that AI agent code execution is confined to controlled environments, preventing access to host systems or external network resources, thereby guarding against malicious code execution, data leaks, or unintended system modifications. For IT services companies like Endava that handle sensitive client codebases, sandbox isolation is not just a technical requirement but a compliance necessity — particularly under the GDPR (General Data Protection Regulation) framework, where the boundaries of client data processing must be clearly auditable. Additionally, sandbox environments support parallel instantiation, enabling Codex to handle multiple independent tasks simultaneously — a crucial infrastructure guarantee for achieving the "weeks to hours" efficiency leap.

Core Use Case: Revolutionary Acceleration of Requirements Analysis

Pain Points of Traditional Requirements Analysis

In traditional software delivery processes, requirements analysis is one of the most time-consuming phases. Business analysts must communicate repeatedly with multiple stakeholders, sort through business logic, write requirements documents, and go through multiple rounds of review and revision. This process typically takes weeks and is prone to misunderstandings and information gaps.

Requirements Engineering has long been recognized as the primary root cause of software project failure. The Standish Group's CHAOS Report shows that approximately 31% of software projects are canceled due to unclear requirements, over 50% of projects experience severe requirements change issues, and the resulting rework costs account for 40%-50% of total project costs. Traditional requirements analysis relies on business analysts (BAs) collecting and organizing requirements through interviews, workshops, and document reviews — a process that is not only time-consuming (typically accounting for 15%-25% of the project lifecycle) but also highly dependent on individual experience and communication skills, making it susceptible to subjective bias. The core value of AI agent intervention in requirements analysis lies in: making tacit knowledge explicit, structuring unstructured information, and parallelizing sequential review processes — fundamentally resolving the efficiency bottleneck at the requirements stage.

The New AI Agent-Driven Requirements Analysis Model

Endava has used Codex to fundamentally restructure this process. AI agents can:

• Automatically parse business requirements: Extract key requirement points from unstructured business descriptions
• Generate structured documents: Automatically produce requirements specifications in standardized formats
• Identify potential conflicts: Cross-validate requirements to discover logical contradictions
• Accelerate iterative feedback: Rapidly generate prototypes or pseudocode to help stakeholders intuitively understand requirements

This approach compresses the requirements analysis cycle from weeks to hours while improving the completeness and consistency of requirements documentation.

Impact on the End-to-End Software Delivery Pipeline

The Multiplier Effect on Development Efficiency

Accelerating requirements analysis is just the starting point. When high-quality requirements documents can be produced rapidly, downstream phases — design, coding, testing — all benefit accordingly. Codex agents can generate code scaffolding directly from well-defined requirements, write unit tests, and even conduct preliminary code reviews, creating end-to-end acceleration from requirements to delivery.

A New Paradigm for Human-AI Collaboration

Notably, Endava's approach does not aim to fully replace human engineers with AI. Instead, it establishes an entirely new model of human-AI collaboration:

• AI agents handle highly repetitive tasks with clear rules
• Human engineers focus on creative decision-making, architecture design, and complex problem judgment

This division of labor frees engineers from tedious operational work, allowing them to invest their energy in higher-value activities.

Industry Implications and Future Outlook

Profound Impact on the IT Services Industry

As a global IT services company, Endava's practice carries significant industry-wide implications. The global IT services industry has long relied on "Labor Arbitrage" as its core business model — deploying large-scale engineering teams in low-cost regions to provide price-competitive services. Indian giants like Tata Consultancy Services (TCS), Infosys, and Wipro have all built organizations of hundreds of thousands of employees based on this model. However, the rise of AI agents is fundamentally disrupting this logic: when AI can complete large volumes of standardized programming tasks at extremely low marginal cost, the competitive advantage of workforce scale will shrink dramatically. As a mid-sized, European-background IT services provider, Endava's agentic transformation strategy carries clear forward-looking significance — by being among the first to build AI agent orchestration capabilities, it positions itself advantageously amid structural industry reorganization. A Goldman Sachs research report predicts that AI could reduce global software development positions by approximately 30%, while simultaneously creating new roles such as "AI Agent Architect" and "Prompt Engineer." The industry's talent structure will undergo profound reshaping. In the future, IT services companies' competitiveness will increasingly depend on their ability to build and orchestrate AI agents, rather than sheer headcount.

The Universal Applicability of the Agentic Organization Concept

While Endava's case focuses on software delivery, the concept of the Agentic Organization has broad applicability. Any organization involving complex knowledge workflows — from consulting firms to financial institutions, from healthcare to manufacturing — can draw on this model to restructure core business processes through AI agents.

Conclusion

The Endava-Codex collaboration showcases a clear path for AI agent technology evolving from "assistive tool" to "organizational infrastructure." When enterprises stop viewing AI merely as an efficiency tool and instead redesign organizational structures and workflows around AI agents, the productivity potential unleashed will represent an order-of-magnitude improvement. This is not just a victory for technology — it is a profound transformation in organizational management thinking.

Key Takeaways

• Endava leverages OpenAI Codex to build an Agentic Organization, compressing requirements analysis from weeks to hours
• An Agentic Organization is not about simply embedding AI tools — it redesigns how organizations operate around AI agents, with theoretical roots traceable to BPR and transaction cost theory
• The new-generation Codex, built on the o3 model, autonomously executes tasks in cloud sandboxes, marking the evolution of AI programming tools from "assisted completion" to "autonomous delivery"
• Requirements engineering is the primary root cause of software project failure; AI agents fundamentally address this bottleneck through structuring and parallelization
• AI handles repetitive tasks while humans focus on creative decisions and architecture design, forming a new human-AI collaboration paradigm
• IT services industry competitiveness is shifting from workforce scale to AI agent building and orchestration capabilities, with structural industry reorganization already underway