Enterprise AI Agents Face Readiness Gap, Endava Shows Path

The Change

Recent research from MIT Technology Review AI shows that 85% of organizations aim to become "agentic" within three years, yet 76% admit their current people, processes, and workflows cannot support that ambition. The gap is not a shortage of models; it is a shortfall in the organizational scaffolding that lets those models act autonomously.

At the same time, OpenAI’s blog details how Endava has re‑engineered its delivery pipeline with Codex, turning a weeks‑long requirements phase into a matter of hours. The contrast highlights a widening divide between vision and execution.

Why Now

Enterprise AI agents have moved from proof‑of‑concept to core business functions in the last 12 months. Google’s I/O 2026 keynote emphasized that Gemini‑powered agents are designed to help workers get more done, signaling that the market expects agents to be production‑ready.

Yet the MIT study warns that most companies are still stuck in legacy silos. The same report cites a “sticky” mismatch between rapid model releases and the slower evolution of governance, data pipelines, and permission frameworks. VentureBeat’s recent piece reinforces this point, arguing that the real bottleneck is not model performance but the permissions architecture that governs what an agent can access.

How It Works

Endava’s approach, as described by OpenAI, centers on Codex acting as a “software‑delivery co‑pilot.” By embedding the model directly into the CI/CD workflow, Codex translates high‑level specifications into code snippets, runs automated tests, and opens pull requests without human hand‑off. The result is a reduction of the requirements analysis timeline from weeks to hours.

In contrast, the typical enterprise that fails the MIT readiness test relies on ad‑hoc scripts and manual approvals. Permissions are locked behind multiple layers of legacy IAM policies, forcing agents to request human overrides for almost every action. Without a unified permission layer, agents cannot execute end‑to‑end tasks, no matter how sophisticated the underlying model.

Bridging the gap therefore requires three coordinated moves:

• People: Upskill staff to design, monitor, and intervene in agentic workflows.
• Process: Redefine hand‑off points so that agents can own entire micro‑services rather than isolated functions.
• Permissions: Deploy a dynamic, policy‑driven access control system that lets agents request, receive, and audit privileges in real time.

Endava’s Codex integration shows a concrete example of aligning all three pillars. The company re‑structured its delivery teams around “agentic pods” that own a full product slice, granting Codex scoped permissions to the codebase, test environment, and deployment pipeline.

Who Benefits

Enterprises that overhaul their organizational design stand to gain faster time‑to‑market, lower engineering overhead, and more consistent compliance. Software firms can shave weeks off the backlog grooming stage, freeing product managers to focus on strategy.

Companies in regulated industries—finance, healthcare, and energy—will especially benefit from a permission framework that logs every autonomous action, satisfying auditors while still allowing agents to act.

Finally, the workforce itself gains a clearer role. Instead of guarding gate‑keeping functions, engineers become supervisors of agentic output, reviewing suggestions rather than drafting every line of code.