How to Scale AI Coding for Real Business Outcomes | Jellyfish Use Case | Sonar Summit 2026

Understanding AI Transformation at Scale

Nicholas Arcolano, head of research at Jellyfish, presented compelling data-driven insights into how organizations are adopting and scaling AI coding tools across their software development lifecycles. At the Sonar Summit 2026, Arcolano addressed the critical questions facing both AI-native startups and established companies attempting digital transformation: what does successful AI adoption look like, what productivity gains should organizations expect, and what unforeseen side effects might emerge from these changes. Jellyfish's engineering intelligence platform provides observability into AI tool adoption patterns, enabling organizations to move beyond surface-level coding metrics toward measuring genuine business outcomes.

Data-Driven Insights from 200,000 Developers

The research presented draws from an extensive dataset encompassing 20 million pull requests across approximately 200,000 developers from around 1,000 companies. Jellyfish combines multiple data sources—including behavioral signals from AI coding tools like Copilot, Cursor, and Claude Code, autonomous agents such as Codex, source control platforms, and task management systems—to create a comprehensive view of AI adoption patterns. This multi-layered approach allows organizations to understand not just how tools are being used, but how those usage patterns correlate with actual code quality, shipping velocity, and business outcomes.

Adoption Metrics: From Weekly Users to Frequent Adopters

Current adoption data reveals steady growth in AI coding tool usage across enterprises. As of January 2026, the median company has approximately 63% of its engineers using AI coding tools on a weekly basis. However, a more predictive metric focuses on frequent users—engineers leveraging these tools three or more days per week—as this habit formation indicates deeper integration into workflows. More than half of surveyed companies now have 50% or more of their engineers as frequent AI users, though significant barriers remain for many organizations. Technical, organizational, privacy, security, and cultural challenges continue to impede broader adoption, particularly in large enterprises and high-compliance industries where rollout proves more complex.

The Rise of Autonomous Coding Agents

A particularly striking finding emerged regarding autonomous coding agents, which represent a significant evolution beyond interactive tools. While less than 1% of AI coding activity involved fully autonomous agents one year prior, the landscape has shifted dramatically. As of January 2026, the median company shows approximately 1% of pull requests completed entirely by autonomous agents—systems that receive specifications or prompts and execute complete coding tasks with minimal human intervention beyond approval. More notably, the 90th percentile of adopting organizations have reached approximately 10% of their PRs completed by autonomous agents, demonstrating exponential growth and indicating that leading organizations are rapidly pulling ahead in their use of this advanced capability.

Key Takeaways

• Median AI adoption has reached 63% weekly active users, with more than half of companies achieving frequent user adoption (50%+), indicating mainstream rather than fringe technology adoption
• Frequent user metrics are more predictive of outcomes than simple weekly active user counts, as they indicate habit formation and deeper workflow integration
• Autonomous coding agents are experiencing exponential growth among leading organizations, with top adopters approaching 10% of pull requests completed by agents—a dramatic acceleration from near-zero usage one year prior
• Organizational barriers remain significant, particularly for enterprises with compliance requirements, suggesting adoption curves will continue to vary substantially across company types and industries
• Data-driven measurement is essential for AI transformation success, combining signals from multiple tools and platforms to move beyond surface metrics toward understanding genuine business impact