Software leaders struggle to maintain control as artificial intelligence generates majority of code
November 11, 2026 · 3 min read
Artificial intelligence now generates 42 percent of code committed to repositories, with that figure expected to reach 65 percent within a year, according to a development survey by Sonar based on responses from over 1,100 professional developers. Seventy-two percent of developers who have tried artificial intelligence coding tools now use them daily, yet 96 percent of developers do not fully trust the code, and only 48 percent always review it before committing. Thirty-eight percent of developers reported that reviewing artificial intelligence code requires more work than analyzing code written by human colleagues.
Tim Chilton, chief advisor on geospatial matters at the British mapping authority Ordnance Survey, acknowledged the pattern. He stated that developers often spend more time fixing code created by artificial intelligence than the reverse, though his organization has not observed this pattern internally. Chilton said his organization is cautious about how developers use artificial intelligence coding tools, and conversations with other digital leaders suggest a similar approach.
Benoît Dageville, co-founder and product chief at Snowflake, described two perspectives on artificial intelligence driven coding: the frightening way, where artificial intelligence takes over tasks without human involvement, and the efficient way, where artificial intelligence helps accomplish the same work at a fraction of the cost. Steve Lucas, chief executive of integration technology specialist Boomi, noted increased production from artificial intelligence tools. Research indicates that 4 percent of public commits on GitHub are already created by Claude Code, a figure expected to rise to 20 percent of all daily commits by year end. Boomi personnel write ten times as much code today as two years ago, Lucas said.
At Ordnance Survey, Chilton uses Coco, Snowflake's coding agent, which handles up to 90 percent of a development task he would otherwise need to complete from scratch. However, the tool's influence is carefully limited. The agent does not work in isolation, and Chilton reviews results in detail. He emphasized that all organizations need a similar process and should value experienced professionals who can use artificial intelligence coding tools effectively.
Gill Haus, chief information officer at Chase, the consumer banking division of JPMorgan Chase & Co, stated that benefits of artificial intelligence coding can only be realized when agents are deployed with strong focus on automating supporting technology and processes, reducing the need for excessive human involvement. Chase never releases services to production that do not meet the company's high standards for quality and security, Haus said.
François-Xavier Pierrel, group chief executive for data and advertising technology at French television channel TF1, emphasized that maintaining control as artificial intelligence writes more code means keeping humans involved in clearly defined ways. His team documents agent-based processes using Atlassian's project management and collaboration tools. Pierrel stated that policy management is crucial as agent-based transformation accelerates, and smart companies establish policies clearly stating when staff may explore an agent, delegate work to it, or release it to production.
Kirsty Roth, chief operating officer at Thomson Reuters, said digital leaders must be proactive. Her company's exploration of models and agents suggests artificial intelligence can facilitate time-consuming coding processes. Roth raised security concerns, noting that Thomson Reuters allocates significant resources to address threats in the artificial intelligence era. She compared the situation to previous major technology trends including digitalization, cloud services, mobile technology, and social media, which took several years to gain traction after initial hype.
Tal Carmi, chief information officer at technology company Walk Me, noted that artificial intelligence coding tools make it much easier for non-information technology personnel to create coded applications. Effective results depend on careful balance between enabling and guidance. Walk Me's technology specialists ask important questions to ensure results are secure, considering login credentials and potentially problematic packages.
Emmanuel Frenehard, chief digital officer at biopharmaceutical giant Sanofi, emphasized that digital leaders must help professionals feel secure. He expressed concern for professionals not using artificial intelligence for coding, saying they might lack jobs in five years. Sanofi guides, coaches, and trains staff, but each person must choose to adopt the technology, Frenehard said.