# Why Humans Can't Keep Up With AI-Written Code - Tariq Shaukat, Sonar **URL:** \ **Category:** Content Creators\ **Tags:** humans-in-the-loop, agents, ai-engineer, automation, coding\ **Created:** [29 October 2026 19:00 UTC](https://www.artofsm.art/t/why-humans-cant-keep-up-with-ai-written-code-tariq-shaukat-sonar/27123 "2026-10-29T19:00:16Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![artesia](https://www.artofsm.art/user_avatar/www.artofsm.art/artesia/32/36_2.png) [@artesia](https://www.artofsm.art/u/artesia)\ **Post date:** [29 October 2026 19:00 UTC](https://www.artofsm.art/t/why-humans-cant-keep-up-with-ai-written-code-tariq-shaukat-sonar/27123/1 "2026-10-29T19:00:16Z")
[![](https://www.artofsm.art/uploads/default/original/3X/6/1/610403d35c6a4f363b15d5be527173a701df489d.jpeg "Why Humans Can't Keep Up With AI-Written Code - Tariq Shaukat, Sonar") ](https://www.youtube.com/watch?v=LQW_Wvqhp04) Tariq Shaukat explains that while AI has dramatically increased code production, human oversight and traditional development methods struggle to keep up, leading to challenges in quality, security, and deployment. He advocates for a new approach centered on the “code confidence formula”—combining clear intent, guiding AI agents, thorough verification, and auditable evidence—to build trustworthy, scalable software factories that can effectively manage AI-generated code. ---
**Author:** ![artesia](https://www.artofsm.art/user_avatar/www.artofsm.art/artesia/32/36_2.png) [@artesia](https://www.artofsm.art/u/artesia)\ **Post date:** [29 October 2026 19:25 UTC](https://www.artofsm.art/t/why-humans-cant-keep-up-with-ai-written-code-tariq-shaukat-sonar/27123/3 "2026-10-29T19:25:02Z")
Tariq Shaukat, CEO of Sonar, opens by drawing a parallel between modern software development and the evolution of manufacturing factories, particularly referencing the Toyota production system from the 1980s. He highlights how Toyota revolutionized car manufacturing by prioritizing quality over volume, fixing systemic issues rather than symptoms, designing for buildability, solving problems immediately, and empowering all workers to stop the production line if needed. These principles have influenced contemporary software development methodologies such as Agile, Kanban, continuous integration, and error budgets, which have helped build effective software factories. However, Shaukat points out that the rise of AI agents is disrupting this established software development paradigm. Recent research shows an explosive increase in code production—up to 2,450% more lines of code and 240% more commits—but this surge has not translated into a proportional increase in software releases, indicating a bottleneck in scaling from code creation to deployment. Many organizations are exploring how to build “agentic software factories” that can automate and manage this new volume of AI-generated code, but challenges remain, especially around security risks and governance. He identifies five key challenges in integrating AI agents into software development. First, human oversight (“human in the loop”) does not scale effectively given the massive increase in code volume. Second, AI agents struggle with understanding and working within large, complex codebases, leading to decreased accuracy as code size grows. Third, errors compound over long tasks, reducing overall reliability. Fourth, large language models (LLMs) are inconsistent and can change their judgments unpredictably, making them unreliable as sole arbiters of code quality. Lastly, the rapid pace of AI-generated code leaves little time to respond to vulnerabilities, emphasizing the need for built-in prevention mechanisms. To address these challenges, Shaukat introduces the “code confidence formula,” which consists of four essential components: clear intent, guide rails (not just guard rails) to steer AI agents, thorough and independent verification of code, and producing consistent, auditable evidence of quality. He stresses that missing any one of these elements results in zero confidence in the software output. Sonar’s products, such as Sonar Vortex and SonarQube, embody these principles by providing intent guidance, reducing token costs, minimizing issues in commits, and delivering robust verification that improves security, time to market, and reliability. In conclusion, Shaukat emphasizes that while AI has made code creation abundant, confidence in that code remains scarce. The future of software development lies in building software factories that not only generate code but also engineer confidence through rigorous governance, verification, and evidence. By adopting these principles, organizations can harness the power of AI agents effectively and safely, ensuring high-quality software delivery in an increasingly complex and fast-paced environment. ## Useful Links - [SonarSource Official Website](https://www.sonarsource.com/) — Central to the video as it presents the company’s solutions for building agentic software factories with confidence. - [Research on AI Agents Impact on Software Development](https://www.youtube.com/watch?v=DDAp9ilhClQ) — Supports the claim about the explosion of AI-generated code and the bottleneck in scaling releases. - [PayPal AI Research on LLM Judge Reliability](https://arxiv.org/html/2610.12083v1) — Supports the claim about the unreliability of LLMs as sole arbiters of code quality.