Jyoti Bansal: Why AI Coding Doesn't Accelerate Software Delivery

The Software Delivery Bottleneck

• 💡 AI coding generates more code but doesn't speed up shipping due to bottlenecks in testing, deployment, and compliance.
• 🎯 Harness focuses on the "outer loop" of software engineering, which involves delivery processes and consumes 60-70% of development time.
• 📌 This outer loop includes numerous workflows like testing, security scans, compliance, deployments, and cost optimization.

Harness's Agentic AI Approach

• 🚀 Harness, since 2017, has used AI/ML for continuous delivery, evolving from neural nets to agentic AI with LLMs.
• 🤖 The platform offers a library of specialized AI agents (DevOps, SRE, AppSec, FinOps) that operate behind a unified "Harness AI" interface.
• ✅ These agents primarily create production deployment pipelines and assist in troubleshooting, ensuring human oversight and auditability, rather than directly deploying to production.

Prioritizing AI Solutions

• 🔑 Initial agent development prioritized pipeline creation, testing (e.g., resilience, end-to-end), and security vulnerability management.
• 🔬 Harness acquired Quieti (formerly ShiftLeft) for its code property graph technology to identify and prioritize exploitable security vulnerabilities.
• 💡 AI costs are integrated into Harness's consumption-based pricing for modules, avoiding separate charges for AI features.

Impact and Future Outlook

• 📈 There's high adoption of AI integration even in large enterprises, often leveraging Harness's Multi-Cloud Platform (MCP) for flexible toolchain integration.
• 🧠 Internally, Harness developers use AI for over 40% of code generation, and leadership uses it for product management and spec writing.
• 💬 Jyoti Bansal views the current AI market as a "bubble" similar to the dot-com era, expecting short-term volatility but significant long-term transformation.