The CIO agenda for agentic
software development at scale

Intro

CIOs have to make tough calls every day about which projects get prioritized based on time, resources and ROI. On one side, core applications need to be maintained, integrated, and upgraded. At the same time, IT teams are expected to modernize systems, reduce technical debt, and accelerate AI. There never seem to be enough people and time.

Agentic software development could radically change the resources required for mission-critical projects, but it requires CIOs to change the organization around it.

We saw this ourselves at EXL when an experienced developer working on an insurance solution with a coding agent completed a six-to-nine-month modernization project in six hours. Hardly an isolated breakthrough, this followed a year of strategic, deliberate work to retrain developers, test tools, define standards and controls, and redesign engineering workflows.

40% of US companies say they’ve fully deployed agentic AI in at least one business function, according to EXL’s 2026 Enterprise AI study, and 56% have made significant changes to their operating models. The companies getting the most value are changing the workflows and the organization around them. Readying the organization is central to today’s CIO agenda.

The developer’s role changes first

In a traditional agile or lean software development lifecycle, people perform most of the work from requirements through testing. Product owners build the backlog. Teams convert requirements into stories. Developers write the code. Testers identify defects. Project managers track progress and coordinate handoffs.

In an agentic process, agents drive each stage. They draft business requirements, create tickets, generate code, write and run test scripts, review code, identify defects, and update Jira before passing the work to the next agent or person.

Early in the process, people may perform most of the work while agents handle defined tasks. As the team learns where agents improve speed and quality without creating unacceptable risk, more work can move to them. A developer who understands the application, its dependencies, its business rules, and its common failure points can give an agent the right context, divide the work appropriately, and judge whether the output is usable.

When the result is wrong, that developer can determine whether the problem came from the request, the prompt, the agent skill, the source material, or the control used to review the output. Coding becomes a smaller part of the role. Application knowledge, problem definition, review, testing, security, and evaluation become more important. Senior developers become more valuable because they can identify weaknesses in the output that a less experienced developer may miss. A junior developer may need more iterations, consume more tokens, and introduce more risk.

Management has to change with the work

Managers need to evaluate developers differently when agents are writing code and closing tickets. More important is whether the team made good decisions and delivered something valuable. Did outputs meet quality standards and drive business outcomes? If problems were found, were they corrected at the source?

Token economics will become an increasing focus and at least part of assessing people and performance. We’re already seeing backlash against “tokenmaxxing.” In addition to the cost, it runs counter the broader shift in CXO focus AI pilots and experimentation to outcomes and ROI. Teams need to use the right data, context, and model for a task. Experienced developers have an advantage because they can use models more efficiently and recognize weak outputs sooner. At EXL, we build that discipline into developer training, workflow design, measurement, and ongoing optimization.

CIOs will also need to manage agents as part of the team by defining their permitted actions, information sources, operating standards, and requirements for human approval. This requires engineering leaders to understand the process in detail. Our senior leaders joined coding sessions to see where the tools were useful, where developers needed more training, and where standards or controls were incomplete.

Quality is crucial throughout

Agentic development increases the amount of work a team can produce, but it can also generate bad work quickly if quality controls and governance fail to evolve.

Within our insurance solution business, we’ve built agents for code review, testing, defect detection, and reporting alongside the agents that generate code. We connected them to Jira so their findings entered the normal development process.

Engineers and technical leads who own the application own the agents, standards, and controls and they’re accountable for the results. When a human reviewer finds a defect, the team looks at the agent that produced the problem and the one that failed to catch it. The team then updates the relevant prompt, skill, standard, or review process so the same issue is less likely to recur.

We also assess outputs against defined requirements for correctness, hallucination risk, prompt injection, token use, and PII handling. Work that falls below a required threshold stops until the cause is known.

Scaling agentic workflows

The best advice for a CIO interested in evolving toward an agentic development lifecycle is not to try to boil the ocean. A CIO can test the approach with a single product, platform, or category of backlog where the team already knows how long the work would normally take.

Preparation starts before tools are widely introduced. Developers first need training in how models work, how to refine agent skills, how to review outputs, and how their responsibilities will change. Standards and controls need to be defined before agents begin producing work at scale. Managers need new performance measures and direct experience with the process they will oversee.

We retrained about 90% of our LifePRO developers over the course of a year. Some were concerned about agents taking over parts of their roles. Instead of avoiding the issue, we showed them how their work would change.

Agentic development can help companies complete long-delayed modernization and maintenance projects and reduce technical debt. How much companies gain will depend on the roles, management practices, quality controls, and governance that CIOs establish around agentic development.