AI agents are not just making developers more productive, they’re transforming the way developers are using AI to build software.
According to Emilio Salvador, vice president of strategy and developer relations at GitLab, the first wave of AI capabilities for developers, like GitHub Copilot or GitLab Duo, were reactive tools for helping developers do tasks like code completion, explanation, or refactoring.
“In these cases, those add-ons were very well defined,” Salvador said during a recent episode of the What the Dev podcast. “They were constrained to specific workflows, and they were able to be very effective, but always reactive and under human supervision all the time.”
He went on to explain that what we’re seeing with agents, along with improvements in generative AI and reasoning AI, is that they’re able to be proactive and take on more complex tasks—in some cases even making decisions on their own.
“It will be up to the developer to decide when to use those agents to take tasks that in the past would have taken months, and they will happen in the background. And when those tasks are completed, the human or the developer will be able to see the final output,” he said.
According to Salvador, the transition from using reactive AI tools to agents is a step-by-step process, so it’s not necessarily a big transition for developers to deal with.
He recommends development teams start with small low-risk projects. For instance, he’s seen a lot of success with small teams using agents for prototyping and proof of concepts. These are tasks where you don’t need high quality results, but you do need something quickly.
For example, recently, Gerry Tan, the CEO of the startup accelerator Y Combinator, said that about a quarter of the current startups in their program have around 95% of their code written by AI.
“That sounds a little scary, but on the other hand, what that means for founders is that you don’t need a team of 50 or 100 engineers,” Tan told CNBC. “You don’t have to raise as much. The capital goes much longer.”
Salvador said, “in those cases, that’s a fantastic example. You have an idea, you need to go to market with something quickly. You need a proof of concept to validate and iterate on. Those are the ideal places for teams to start with, to evaluate the capabilities and also to what extent they can be used in their context.”
Of course, it’s important to keep in mind that “throwing technology at a problem is not going to solve anything,” he said. Development teams need to be strategic about how they use these technologies. Salvador said that AI is an amazing tool, but it can be misused too, so teams need to be defining a strategy and taking it one step at a time to be successful.
He also recommends organizations remember that humans are the limiting factor in any of these projects. “We are all humans. We need to adopt our technology and understand and embrace the value that it brings. And I think that’s why, like in any other when you think about embracing or adopting a new technology, that change management process is always underestimated.”
His advice would be to start building, identify the technologies you want to use, find champions within your organization that understand and can communicate the value to others, and have a clear sense of direction on how you want to use these technologies.
