“The harness is where the hard work is”: Harness bets on agents that enterprises can trust in production

Frederic Lardinois (0:02)

Jyoti, thanks for being here again. It's the last time we talked, I think it was in October of last year.

At the time, you were already announcing a lot of kind of AI features and you're announcing more AI features today, obviously, because that's the world we live in now. But it was interesting before we talked, your PR representative mentioned, we are in the fourth phase of harness. And I was just wondering, what phases have we been through and how do we get to this point where harness has become an AI company to some degree?

Jyoti Bansal (0:37)

Yeah.

It's interesting that how harness has evolved over the last, since we started and the name harness, when I started the company, I gave the name harness because I thought all the code that the developers are writing, it needs a safety harness around it, and we are the ones who can tie it all together on all the downstream processes that happen after coding, from deployments to testing, security, governance, everything.

And how come the industry stand in name for, how do you govern all the agents and everything, how do you tie it together? So, I feel like fortunate in our name of the company. But when we started harness, let's talking about the four phases here. The phase one, we started with, the biggest gap in DevOps was reliable production deployments, like how do you do production deployments in a reliable way? So, CD part of the CI CD, a lot of the systems were built for builds and testing, but not for automating the deployment processes. So, that's where we started with, we built one of the best platforms for automating the deployment processes. We always had notions of AI from day one, even before LLMs were out there. So, not using LLMs, but for some of the harder problems around how do you do a deployment verification in an automated kind of way, how do you do a kinetic deployment in an automated way, all of those things, we have been doing a lot of AI since the very beginning. So, that was kind of our phase one. In phase two, we started realizing the problems which is our result always, which is so we actually from deploying everything else, everything that happens after coding, which is like building the CI part of the job, testing, resiliency testing, security testing, very defocus on security, because if your security teams are disjointed from your DevOps, then you cannot really ship to production in an automated kind of fashion, cost optimization, all aspects of the process. So the way we always looked at developers write code, that's normally about the first 30, 40% of the job. And then you have to ship that code to production. And you have to do it in a safe, reliable way. And so that's like the 60, 70% of the jobs, all the tasks that you would need to do. We started bringing some of the best capabilities into doing it. So that was our phase two. Phase three for us was how do we bring AI to drive most of those. So over the last couple of years, we have been building what we call like the expert agents. These agents are the ones that will help you create things. Like unless you want to set up a CI CD pipeline for your organization. Also, our expert agent can do this for you in minutes. But expert agents that are rooted into what we call a software delivery knowledge graph, which is a knowledge graph of everything that happens in your organization from your security policies, your infrastructure, your services, how your teams operate. All of those aspects like coming together into a context engine that goes into the agents and the agents can help you set up your pipelines, your security testing, your resilience testing, your cost optimization, all the things that you want to set up in your DevSecOps that our expert agents will do. So that was our phase three. Today, actually, I'm very excited about the phase four, which is evolution of the pipeline itself in a significant way.

If you look at how a CI CD pipeline works, or any kind of a delivery pipeline works, there's normally a pipeline orchestration process like a workflow which harnesses provided one of the best workflow engine around these delivery pipelines for many years now. But the steps in the workflow are normally deterministic steps. They're deterministic steps of like, do a community's deployment, review these logs, run the security test, all kinds of deterministic steps that you normally do. And these are pipelines that are deploying into production and some of the most complex applications in the world, for example. But now what we are launching today is worker agents or autonomous worker agents. These autonomous worker agents becomes the steps in the pipeline. So instead of having scripts in the pipeline or deterministic steps, you can put agents that can reason in the inside the pipeline to do things.