EY has achieved 4x to 5x productivity gains in software development by integrating AI coding agents directly into its internal engineering standards, compliance frameworks, and code repositories, according to Stephen Newman, EY Global CTO Engineering Leader.
The gains did not come from simply deploying a tool. Newman’s team spent 18 to 24 months building the cultural and technical foundations required to make semi-autonomous coding work at scale across EY’s suite of audit, tax, and financial platforms.
The Core Problem With Coding Agents
Coding agents can produce thousands of lines of code in minutes. But volume without context is a liability. “You can generate a ton of code, but it doesn’t mean really anything, right? It’s got to be code that is integratable, that is compliant, and you don’t want to create more work on the back end just because you sped up the code generation process on the front end,” Newman said.
Without access to EY’s internal repositories, engineering standards, and source catalogs, agents produced generic output requiring extensive rework. Newman calls this accessible internal context the “context universe.” Connecting agents to it was the technical breakthrough that unlocked deployable code at scale.
Organic Adoption Over Mandated Rollout
EY started with GitHub Copilot-style assistive tools, giving engineers time to build comfort with prompt engineering before introducing more autonomous systems. Newman was deliberate about the adoption model. “It’s important to bring AI capabilities as a ground-up organic adoption rather than force them onto the users,” he said.
Productivity gains plateaued once developers moved past basic code generation. The next step required deeper integration, pushing developers toward building, deployment, and operationalization workflows supported by agents with full context access.
Choosing the Right Agent Platform
EY evaluated three agent platforms: Lovable, Replit, and Factory’s IDE-based Droids. Rather than mandate a single tool, Newman’s team tracked adoption, usage, and productivity across all three, letting developer preference guide the decision.
Factory emerged as the clear choice. Adoption “took off like wildfire” once it moved from evaluation to pilot. Demand grew fast enough that EY had to throttle traffic to Factory and Droids and restrict repository access until compliance and security approvals were in place.
What Agents Handle — and What They Don’t
EY built a workload classification framework to determine when to delegate to agents and when to keep humans in control. High-autonomy tasks delegated to agents include:
- Code review
- Documentation
- Defect fixing
- Greenfield feature development
Tasks that still require human oversight include large-scale refactors, architecture decisions, and cross-system integrations. The distinction reflects where agents produce reliable output versus where errors carry compounding downstream risk.
Developer roles shifted accordingly. Engineers moved from writing code directly to orchestrating agents, directing them toward the correct databases and repositories rather than executing tasks themselves.
Measured Gains
During the early adoption phase, EY measured efficiency gains ranging from 15% to 60% across different developer personas. The 4x to 5x productivity figure reflects the more mature phase Newman describes as “horizon model development,” where semi-autonomous agent execution runs at scale with a team of orchestrators rather than individual coders.
Newman acknowledged the difficulty of attributing gains to any single factor. The productivity improvements came from a combination of trial and error, technical integration, and sustained behavioral change within developer teams.
Photo by Irvan Smith on Unsplash
This article is a curated summary based on third-party sources. Source: Read the original article
