Endava, a global software services firm, has rebuilt how it ships software around AI agents. According to OpenAI, the company is using ChatGPT Enterprise and Codex to speed up delivery, automate routine workflows, and grow an AI-native culture across the whole organization. What stands out here is the scope. This isn’t one team running an experiment. It’s a delivery model redesigned from the ground up.
The Endava case is a useful template if you want to do the same. Here’s how to approach it, step by step.
Quick Start
What you’ll learn: how to redesign software delivery around AI agents, the way Endava did with ChatGPT Enterprise and Codex.
What you need: an enterprise-grade AI assistant (Endava chose ChatGPT Enterprise), a coding agent like Codex, leadership buy-in, and a few teams willing to go first.
Step 1: Start with a secure, enterprise foundation
Endava built on ChatGPT Enterprise, not a consumer tool. This matters because enterprise plans give you data controls, admin oversight, and the privacy guarantees a services business needs when client code is involved. Lock down the foundation before you scale usage. Trust comes first, adoption follows.
Step 2: Put a coding agent directly in the workflow
The company brought in Codex to handle real engineering work, not just chat. A coding agent that reads your repos, drafts changes, and runs tasks sits where the work actually happens. The point is to cut the gap between intent and shipped code. Pick the parts of your pipeline where engineers lose the most time and aim the agent there.
Step 3: Automate the workflows nobody enjoys
Endava used agents to automate routine workflows across delivery. Think code review prep, documentation, test scaffolding, status updates, and the glue work that eats hours. This is where you’ll see early wins. Automating dull, repeatable tasks frees senior people for the judgment calls only humans should make.
Step 4: Build an AI-native culture, not a pilot
The biggest move Endava made was cultural. The goal was an AI-native way of working across the enterprise, not a sandbox a few engineers visit. This is the hard part. Tools are easy to buy. Habits are hard to change. Make agent use the default path, share what works between teams, and treat AI fluency as a core skill rather than a bonus.
Step 5: Measure delivery, then expand
The stated aim is to accelerate software delivery. So measure that. Track cycle time, throughput, and how much manual toil drops once agents are in the loop. Prove the gains on a few teams, then roll the pattern out wider. Evidence beats enthusiasm when you’re asking a whole company to change how it works.
Why this matters
Endava sells software delivery as its product. When a services firm rewires its own delivery engine around AI agents, that’s a signal about where the industry is heading. The competitive edge is shifting from how many engineers you can hire to how well your people and your agents work together. Firms that treat AI as a culture, not a feature, will move faster than those still running pilots.
Best practices to keep in mind
- Choose enterprise-grade tools when client or proprietary code is in play. Security is the price of entry.
- Aim agents at your slowest, most repetitive steps first for the clearest payoff.
- Don’t stop at a pilot. The value shows up when AI becomes the default way of working.
- Keep humans on the decisions that need judgment, taste, or accountability.
Next steps beyond this guide
Start small and concrete. Pick one delivery team, give them an enterprise assistant and a coding agent, and set a 90-day target for cycle-time improvement. Document what works and what breaks. Then use that team as your proof point to bring the rest of the org along. Full details on Endava’s approach are available at the original source.