AI Increases Code Production Speed but May Expand Software Vulnerabilities
November 14, 2026 · 2 min read
Executives from Accenture and Tempest have identified productivity gains and security challenges resulting from the adoption of artificial intelligence in software development.
Accenture reports that artificial intelligence provides productivity gains of 40% to 50% across its projects, with certain tasks seeing increases of approximately 70%. These figures reflect faster delivery and higher quality for clients, although the specific methodology for these measurements and the exact activities that yielded the highest results were not detailed.
To achieve these results, Accenture is implementing AI-native squads. These are teams structured to integrate artificial intelligence into development processes from the project conception stage rather than using it only for specific support activities. This strategy utilizes the GenWizard platform, which focuses on applying artificial intelligence throughout the software development cycle using shared and reusable resources to maintain quality and security controls.
The shift toward automation has raised concerns regarding the professional development of future specialists. Mauro Silva of Tempest noted a risk that reducing entry-level opportunities could hinder the ability of junior developers to gain the experience necessary to become senior professionals. This creates uncertainty regarding the future quantity and quality of senior experts. In response, Accenture is developing initiatives to accelerate the training of junior workers and align their skills with client needs by connecting them with the experience of senior professionals.
Increased production speed also introduces security risks. Tempest warns that artificial intelligence does not eliminate vulnerabilities and may increase the volume of problems as more code is produced without proper verification. To mitigate this, the company advocates for security by design, which involves incorporating security from the start of system conception. Recommended measures include segmenting the activities of AI agents, using sandboxes, performing penetration tests, conducting code reviews, and implementing independent verifications to ensure no single tool handles development, review, and validation.