Harness's Jyoti Bansal says CI/CD needs to be reimagined for AI coding
When Harness launched in 2017, it was a different world. Developers wrote code in those days and Harness was designed from the beginning to help build, test and release that code faster. While it's still focused on that, AI agents are producing more code much faster and the company has had to adjust to that shift.
Jyoti Bansal, founder and CEO of the company, appeared recently on the FastForward on PPN podcast, and he discussed how much the company has changed since those early days and how much AI coding has changed how the company works.
"The idea behind Harness when we launched was that the developers are writing all this code, and we need to have a reliable way to deliver and ship the code, and that was the harness, this safety harness for all of that, that will bring it together and bring intelligence to it," Bansal said on the podcast.
It's interesting because harness has come to mean something different since then, the structure you build around a model, but back then this is what Bansal was thinking when he named the company. "I call ourselves the OGs of harness," Bansal joked.
"What happened with AI now is that need has become like multiple orders of magnitude more because the code is coming out so fast," he said. And it's not just having an impact on the developer customers, that dynamic comes into play internally too.
"The number one thing is that the velocity is much higher now than it was a decade ago, for sure. Our team did an estimate, and we are shipping one big feature every 11 hours right now. I could not have imagined that a decade ago," Bansal said.
But the speed brings its own set of problems, and maybe a need for a new approach. The CI/CD pipeline was built for human developers coding at human speed. With AI agents now generating much of that code, the pipeline concept itself has to evolve.
Bansal pointed to an MIT/Wharton study published by the NBER in May which found that while AI increased coding activity by 180%, the amount of code that actually shipped rose just 30%. That's quite a gap and speaks to the speed problem.
"I think the bottleneck is the entire CI/CD pipeline in my mind. If you think of the CI/CD pipeline, it was all designed for human speed," he said. And the problem as he sees it, is there is no way to scale that idea to AI coding speed. "That is the bottleneck right now. So you have to completely reimagine that."
