Coding agents won't replace software engineers: Datadog Cofounder

Datadog CTO Alexis Le Quoc says coding agents can dramatically accelerate software development, but humans remain essential to manage complexity, security and team-level work

Datadog, the US company which provides cloud-based monitoring and data analytics platform, has about 100 employees in India working across sales, solutions, engineering, and marketing. In a virtual interaction from New York, Cofounder and Chief Technology Officer Alexis Le Quoc talks to Avik Das about the complexities of modern software, impact of coding agents and why they are unlikely to make software engineers irrelevant. Edited excerpts:

 

Datadog describes itself as a cloud-based monitoring and analytics platform. What are some of the key parameters that you monitor for enterprises?

 

We started about 16 years ago during the onset of cloud and our initial position was to help companies understand their cloud deployments, as there was a lack of visibility across the board. It is a bit of a similar story with artificial intelligence (AI), where there's a tremendous lack of visibility that's coupled with a tremendous capability that AI brings. And that's causing a lot of anxiety among customers and an opportunity for us to provide that visibility and sense of control on anything that touches performance, security, and cost of AI applications. That makes us a critical piece of the infrastructure. Just like it is very risky to run a large cloud footprint without any kind of observability at all, it is equally risky and costly to run AI applications without observability.

 

Your Cofounder Olivier Pomel has often talked about how modern software has become extremely complex. What are some of the complexities that have emerged because of AI in software?

 

There are a couple of dimensions to this. One is AI, which allows us to create more software and it is not that we have more people who manage that. So at some point, the quantity of software can reach a scale velocity where without the proper tooling, it's too complex and we won't be able to sustain it. The other part is, since software can use AI, the behaviour that the technology brings into traditional software is very different from what we used to see. Software used to be the epitome of deterministic behaviour, where you write the code and know what that is supposed to do.

 

But now, we have these models that are inherently statistical. And so, there are a bunch of techniques where you can generally coerce behaviour to do what you want, but you never have the mathematical certainty. So that does not quite create a deterministic behaviour. It's too complex for the human mind to just observe with your own eyes. At the end of the day, the business is still by and for the people, which means that humans are very much in the loop. However, they cannot be in the loop in the same way as earlier, where one could see what the software was doing.

 

How has AI impacted the coding life cycle?

 

Last year, there was a step improvement in the ability of the coding agents to write software at a complexity that effectively approaches, or sometimes surpasses human ability. For example, we have had projects where we've been able to spend a few hundred dollars on very well-skilled, well-defined problems or software architectures and create systems that would have taken three extremely expert engineers six months to build. That has been done in a couple of weeks with one person and a coding agent. That doesn't mean all of a sudden I need just one engineer rather than 10. Coding agents are really good at taking one small piece of software architecture and making it a lot better. But the reality of software production is it's a multi-team effort with different operating roadmaps, cadence, and priorities. And the coding agents don't make that go away. They exacerbate some of the friction that exists.

 

How do organisations then reshape their workforce with humans and agents working together?

 

With coding agents, we have pockets where small things go very fast and then they hit a wall and slow down. For a while, we have been operating with the two-pizzeria team rule of six to 10 people, and that is now being questioned. It seems unlikely that organisations will be a bunch of individuals who work on their own and there are no teams anymore, because that is not sustainable. So, the industry is figuring out that while we can write a lot more code faster and correctly, can it be done securely. And for that, we need people. I don't see how coding agents are going to obviate the need for software engineers or software engineering as a profession is over.