The human part of AI development

Why this conversation

AI has made code fast to produce, but someone still has to read it, understand it and answer for it. Katherine Druckman of JetBrains has spent years in open source communities, and at the conference she gave a talk about the human side of this work. She was the host’s last conversation of AGNTCon + MCPCon North America, recorded in San Jose on Friday 23 October, the event’s second day. The rest of that Friday is in AGNTCon + MCPCon North America 2026: day two.

Open source, for a long time

The host opened by noting that she does a lot of things. She agrees. She has been part of open source communities for so long that she would rather not count the years. She takes part in the work of the Agentic AI Foundation, of which JetBrains is a member, and JetBrains also belongs to the OpenSSF, the Open Source Security Foundation, where she has been active for a while. What draws her is the work happening in the open on hard problems. Will it shape the future? She hopes so.

The cost that has not fallen

Asked what JetBrains is doing with AI, she says the company is taking AI development very seriously, for open source communities and for developers of every kind. Generating code now costs little time. Reviewing it does not get cheaper, and neither does understanding it or trusting it, and those costs matter more and more.

That raises the question of where the human goes. Writing code is becoming a smaller part of the job across the industry, yet people remain responsible for the code that AI writes. “We have to know what’s in our software.” Developers must answer for it, trust it and understand it. JetBrains has taken that position, she says, and puts itself behind developers of all kinds, open source or not.

What a human in the loop needs

This is a subject she gets a little soapboxy about, and it was the subject of her talk: the human part of the work, and the support the humans doing it need. One of her slides was about human in the loop, a phrase the industry is fond of. It is often described as a kind of safety gate. For her, it is also, in the end, an actual person, the one who is accountable, who needs to understand the code base, and who needs a reason for each change to it and an understanding of that reason. Defining that person is part of what makes the moment interesting.

“What does that human need in terms of support, resources and tools?” Effective tools, first of all. And, back in open source, resources and support, and, to be honest about it, budgets. Budgets can often be directed to the right place, and at other times they turn into time: people pitching in. In an open source community, she says, people give their own time and understanding, and all of these things go together.

AI-first developers, and everyone else

JetBrains, she says, enables the developers who build with AI, and it also enables those who do not; nobody can take for granted that every developer works that way. The company has just launched JetBrains Air, which she encourages everyone to look at. At the conference it also showed something closer to the cutting edge: a tool meant to help developers understand cost, including how many tokens they spend on particular tasks, and to break down organizational knowledge, so that what a development team learns can be shared across the company. Again, she says, it comes back to understanding what is in the software.

“What a time to be a developer.” For a starting point she suggests jetbrains.com. She also invites people to find her and her JetBrains colleagues on social media.

When it comes out

The episode is edited and in the release order. Like every episode it goes out on YouTube first, then on Apple Podcasts, Spotify and Amazon Music.