The wedge nobody talks about
Most companies are still thinking about LLM tooling as autocomplete with better syntax awareness. Endava, a global software contractor with engineers across Europe, the Americas, and Asia, rewrote the playbook: they encoded their senior architects' judgment into Codex and used it to compress requirements analysis from weeks to hours.
They now call themselves an agentic organization—a company where senior expertise is codified into agents that work alongside teams across the entire client engagement lifecycle, from intake to ideation and delivery. This isn't about replacing engineers. It's about distributing senior judgment at scale.
"We went from producing a lot of the code ourselves to now overseeing the work that Codex can produce," says Joe Dunleavy, Endava's regional CTO for Europe. "The quality of output has just gone up exponentially."
Senior judgment, available in parallel
The real unlock at Endava wasn't faster code generation. It was knowledge transfer at scale.
Mike Krolnik, Endava's Global SVP of Agentic Architecture, describes it clearly: "Senior architects like myself, coming from complex environments, are able to articulate what we want, and Codex makes that an accessible piece of information for the more junior people on the team."
In practice, this means junior developers are now doing work that would normally require years of pairing, code review, and mentoring. They're handling architectural decisions with Codex acting as a real-time guide to best practices. "I can give Codex a point of view, and when they're working, it will help them understand this point of view," Krolnik explains. "They can ask questions about things they don't understand."
This is the compounding multiplier: a single senior architect's perspective, once encoded into Codex, can guide multiple teams in parallel. The alternative—traditional mentorship—requires that senior to be in one place at one time.
From weeks to hours: the legal contract case
Endava's own legal team stress-tested this in the wild. They brought engineering a messy problem: thousands of pages of contracts to review against specific criteria.
Translating legal requirements into engineering specifications normally takes weeks of back-and-forth meetings, document revisions, and clarification cycles. Instead, Krolnik's team recorded a two-hour deep-dive with the legal stakeholders, fed the transcript to Codex, and generated a working requirements spec.
What could have taken one to two weeks of revision was compressed into two one-hour meetings and a usable spec. The time savings are real, but the strategic shift is bigger: analysis, design, and build are no longer sequential handoffs between specialists.
"Each of these stages used to take days or weeks of analysis," Krolnik says. "Now with Codex packaging together analysis, design, and build, we can do that as a single unified tool."
Desktop agent, not code assistant
The biggest mistake teams make is treating Codex like GitHub Copilot with a better context window. Endava's breakthrough came when they stopped thinking of it as a coding tool and started treating it as a desktop agent across the entire lifecycle.
"Codex has matured as a tool," Krolnik says. "We use it for requirements analysis, design, specifications, development, and operations; it's a general desktop agent across our whole lifecycle."
That means:
- Generating design documents and architecture diagrams live in client sessions
- Producing specifications from meeting transcripts
- Translating between stakeholder language and engineering requirements
- Guiding operations and deployment decisions
Krolnik describes showing clients real-time visualizations: "You can tell it to draw a diagram of the proposed software architecture so it's easier to understand for our clients. It rapidly accelerates the back-and-forth, and it really opens a lot of doors."
The fastest ROI at Endava came from applying Codex to places where they'd never used a coding tool before. Requirements analysis. Client communication. Design documentation. These aren't coding tasks, but they're where senior judgment matters most.
The playbook for other orgs
Endava's advice to teams just starting is counterintuitive: pick a non-coding workflow first.
Dunleavy puts it bluntly: "The first piece of advice is you need to get beyond thinking about what you want to do and actually get in and try it." Don't start with code generation. Start with requirements analysis, design docs, or client communication—places where your team has never used AI tooling.
The three leadership lessons from Endava:
- Codify your seniors. The largest leverage comes from capturing senior architects' judgment in
Codex, so junior team members get senior guidance as they execute. - Treat Codex as a desktop agent, not a coding assistant. The biggest unlocks came from applying
Codexto requirements, design, client communication, and operations alongside code. - Don't just think about it, really try it. Pick a non-coding workflow first. The fastest way to see
Codex's full value is to use it where you've never used a coding tool before.
Krolnik sums it up: "Codex magnifies every skill I have, and everybody who learns how to use it gets every skill they have magnified."
What this means for AI product teams
Endava's model is a preview of how large professional services firms will integrate LLMs over the next 18 months. The pattern is clear:
Single-agent coding assistants are table stakes. Multi-agent orchestration across the full delivery lifecycle is the differentiator.
The companies that win won't be the ones with the best autocomplete. They'll be the ones that figure out how to bottle senior judgment, distribute it at scale, and collapse the distance between a client conversation and a working spec.
Endava didn't wait for a perfect product. They took Codex, applied it to the highest-leverage non-coding workflows, and encoded their senior architects into the loop. That's the move.
Dunleavy's closing line is the one to remember: "What Codex has really helped us do is have small teams of people deliver massive value in a very condensed timeframe."
That's the wedge. Smaller teams. Bigger outcomes. Senior judgment, available in parallel.