The New Stack Podcast Harness CEO Jyoti Bansal on Why AI Coding Doesn't Help You Ship Faster
18 snips
Oct 10, 2025 Jyoti Bansal, co-founder and CEO of Harness and a seasoned entrepreneur, dives into the complexities of AI in software development. He reveals how AI tools often create bottlenecks in testing and deployment rather than speeding up the process. Bansal discusses the importance of human oversight and the role of specialized agents within Harness AI that enhance pipeline efficiency. He also shares insights on the evolving landscape of AI, likening its current hype to the dot-com bubble, which will ultimately lead to meaningful innovations.
AI Snips
Chapters
Transcript
Episode notes
AI For Delivery Predated LLM Hype
- Harness used AI and ML for delivery problems long before LLMs became mainstream.
- LLMs accelerated capabilities but AI in delivery predates generative models.
Outer Loop Dominates Engineering Time
- Most engineering time at scale is spent in the outer loop (delivery) not inner-loop coding.
- Delivery includes many workflows like testing, security, deployments, and cost controls that consume 60–70% of time.
Break Tasks Into Contextual Agents
- Break delivery tasks into purpose-built agents and pass only the right context to each.
- Build a knowledge graph of your environment so agents can make correct, auditable decisions.